Environmental protection has always been a concern for iGEM, in which biodegradation of PET plastics has become a research hotspot for various teams in recent years. We noticed that plastic-binding peptides could be a potential tool to enhance the degradation efficiency of PETase through their ability to bind to PET plastics, which has not been studied in depth by any team so far. In 2024, we are committed to tackling the existing PET microplastic contamination problem by mining PET-binding peptides with the help of deep learning and constructing PETase-PET-binding peptide fusion proteins through synthetic biology methods. We would like to actively share our experience and results obtained by our team on PET-binding peptides with all the teams, so that we can give some help and guidance to the future iGEM teams, and make the iGEM community more colorful. Our Contribution section will be divided into the following three parts: Dry lab, Wet lab, and Human practices.

First, we constructed a model based on Long Short-Term Memory (LSTM) networks to deeply analyze the local and global dependencies within protein sequences. Compared to traditional Recurrent Neural Networks (RNNs), LSTM introduces gating mechanisms that effectively overcome common issues of vanishing and exploding gradients when processing long sequences, significantly improving the model's ability to capture long-range dependencies in protein sequences. Using the LSTM model, we successfully performed an initial screening of short peptides with potential hydrophobicity. To address LSTM's limitations in handling three-dimensional protein structure information, we further introduced Graph Convolutional Networks (GCNs). By incorporating an attention mechanism, the model adaptively enhanced its ability to recognize and capture key features in the three-dimensional protein structure. We utilized AlphaFold3 to generate three-dimensional protein structure data as input for the GCN model, accurately extracting local features and long-distance interactions within the protein structure. Additionally, we combined one-dimensional amino acid sequences with three-dimensional structure information for multi-layer screening. This process not only preliminarily identified short peptides with high affinity for PET plastic but also eliminated low-affinity candidates through three-dimensional structural information. Ultimately, 480 candidate short peptides were selected, and from these, 16 short peptide sequences were finally chosen based on insolubility index and scoring for experimental validation by the wet lab team. The table below presents the results: You can view detailed information on our Model page.

Table 1  Scoring table of 480 amino acid sequences

Table 2  The 16 target peptides after secondary screening

Our study innovatively combines the strengths of both LSTM and GCN models (Fig. 1), enhancing the dual analysis capability of protein sequence and structural information. This approach not only provides a new method for future iGEM teams to predict PET-binding peptides, but also offers strong technical support for deep learning in the accurate prediction of short peptides with high affinity for PETase and plastics.

Fig. 1   Model general workflow diagram

For further improvement, we used Mutcompute-super to introduce beneficial mutations at specific sites of PET-binding peptide, and then used the extracted bioinformatics features as the input for the 3D deconvolutional neural network of Mutcompute-super. We then combined a deep learning classifier to predict the optimal amino acid mutation combination to optimize the protein's folding conformation, and further improved the degradation ability of the fusion protein towards PET substrate ^[1] (Fig. 2).

Fig. 2   Heatmap of the mutation scoring situation

(a). Mutation scoring situation of the 60-2 short peptide. (b). Mutation scoring situation of the 50-4 short peptide. (c). Mutation scoring situation of the 50-5 short peptide.

The model we used helped us in protein design through amino acid microenvironment based with EMO neural network, you can view the details on our Model page. We hope to recommend this model to future iGEM teams to provide new tools for molecular modification.

For the 16 PET-binding peptides obtained from the dry lab by LSTM and GCN models, we constructed fusion proteins and performed validation characterization to determine the enzymatic properties and degradation ability of each fusion protein (Fig. 3).

Fig. 3   Identification of PET-binding peptides

(a). Degradation of PET microplastics by C-terminal PET-binding peptides with G4S linker. (b). Degradation of PET microplastics by N-terminal PET-binding peptides with G4S linker. (c). Degradation of PET microplastics by C-terminal PET-binding peptides with SLE linker. (d). Degradation of PET microplastics by N-terminal PET-binding peptides with SLE linker.

We obtained peptides 50-4-N, 50-5-C, and 60-2-N that effectively enhanced the PETase degradation efficiency towards PET microplastics. Based on the optimal mutation sites predicted by Mutcompute-super model, we constructed and characterized the corresponding mutants of fusion proteins. Finally, the degradation experiments on the fusion proteins measured a maximum 207.49% increase in product release compared to PETase (BBa_K5157103), achieving quite significant results (Fig. 4).

Fig. 4   The expression and PET degradation results of fusion protein mutants

(a). Degradation of PET microplastics by the mutant 50-4-N-G4S-PETase. (b). Degradation of PET microplastics by the mutant PETase-G4S-50-5-C. (c). Degradation of PET microplastics by the mutant 60-2-N-SLE-PETase.

You can check out the details on our Parts page and Results page.

We believe that the results of our team's work on enhancing PETase substrate binding with PET-binding peptides will provide new ideas for future iGEM teams and provide experience for future iGEMers to carry out binding peptide-related studies.

His tag BBa_K4790077

We adopted a similar plasmid construction method to that of iGEM23_NNU-CHINA, and also added a His tag on our fusion protein to facilitate purification.

However, during the experiment, we found that some constructed fusion proteins were difficult to be purified by Ni²⁺ columns. Instead, we chose to replace the purification method to thermal purification. For proteins with good thermal stability, the purification of target proteins could be achieved as well (Fig. 5).

Fig. 5   SDS-PAGE analysis. M: Marker; FS: Fermentation supernatant; PC: Precipitation of cell fragmentation; SC: Supernatant of cell fragmentation; NP: Ni²⁺ column purified enzyme; TP: Thermally purified enzyme

We are of the opinion that due to the hydrophobicity of the selected PET-binding peptide, it may have affected the protein conformation, resulting in the fusion protein being difficult to bind to Ni²⁺ ions and thus difficult to be purified.

We believe that our findings on the effect of PET-binding peptides on the properties of fusion proteins and the choice of purification of PET-binding peptide-containing fusion proteins will be useful for future iGEM teams to carry out research on binding peptides.

EAAAK linker BBa_K4990005

Linker as a fragment connecting enzyme and short peptide is categorized into flexible linker, rigid linker and in vivo cleavable linker. We tried to replace the linker to optimize the spatial conformation between the enzyme and the PET-binding peptide, reducing the mutual influence between the two components, thus improving the substrate catalytic efficiency of the fusion protein.

We chose the commonly used linker G4S (BBa_K5157012), EAAAK (BBa_K4990005), and newly uploaded 10A (BBa_K5157015), AG (BBa_K5157014), and SLE (BBa_K5157016) from the region of the endogenous spacer sequence region of PETase and Tr (BBa_K5157013) originated from Trichoderma reesei ^[2] as linkers to construct fusion protein.

Molecular dynamics simulations show that the G4S, Tr, and AG linkers are flexible in the fusion protein, while the SLE, EAK, and 10A linkers function as rigid linkers in the fusion protein (Fig. 6).

Fig. 6   Molecular dynamics simulations

(a). RMSF analog. (b). RMSD analog.

We then constructed fusion proteins by replacing flexible linker G4S with flexible linker Tr, AG and rigid linker SLE, EAAAK, 10A, respectively, then measured enzyme activity after fermentation. As shown in Fig. 7, the highest fusion protein activity was constructed via G4S in the flexible linker, while in the rigid linker, it was SLE. Therefore, we choose to replace the flexible linker G4S with the rigid linker SLE for linker optimization.

Fig. 7   Enzyme activity of fusion protein with different linkers

You can view detailed information on our Part page.

We hope that our results of characterizing different linker properties will help other iGEM teams in choosing linkers.

While HP's activities serve our own projects, we also want to leave some resources behind for future iGEM teams to reference and reuse.

Given that there is less public awareness of microplastic pollution, we created popular science handbooks around microplastics, which sparked an enthusiastic response and surprised us with their enthusiasm for learning.

At the same time, we recorded videos about PET degradation and posted them on various social media to expand the knowledge of synthetic biology and microplastics pollution for the public to learn relating knowledge anytime and anywhere through the Internet.

We also actively communicated with other iGEM teams, shared our experience, organized and participated in many conferences, and created group chats on our topics, providing a platform for academic exchange of synthetic biology.

We aim to use these transferable ways to light up more people's interest in synthetic biology and PET microplastic degradation, and promote the sustainable development of synthetic biology and PET microplastic degradation. We hope that this information will remain in the iGEM archive and that iGEM users can download this information if they want to use it.

[1] Deng Z H, Cai C, Wang S T, et al. A protein design method based on amino acid microenvironment and EMO neural network, CN118136092A [P/OL]

[2] Dai L, Qu Y, Huang J W, et al. Enhancing PET hydrolytic enzyme activity by fusion of the cellulose-binding domain of cellobiohydrolase I from Trichoderma reesei [J]. Journal of Biotechnology, 2021, 334: 47-50.