Model

Mutant protein structure prediction using SWISS-MODEL

sWISS-MODEL is a homologous modeling server based on ExPASy web,designed to make protein modeling accessible to all life science researchers worldwide. Homology modeling, also known as comparative modeling, predicts protein structures based on sequence homology with known structures., It is basedon the principle that if two proteins have a high enough sequence similarity, they are likely to havevery similar three-dimensional structures.

lt therefore relles on the identification of one or moreknown protein structures that may resemble the structureof the query sequence, and on the generation of alignments that map residues in the query sequence to residues in the template sequence. Thus, if one of the protein sequences has a known structure, that structure can be copied to an unknown protein with a high degree of confidence.

The sequence and structure of PETase have been reported, from which we predicted the structure of the mutant using homologous modeling. The result was shown in Figure 1. The mutated residues were 87, 121, 159, 186, 238, 242, 246 and 280, respectively. The Laplace diagram (Figure 1C) shows the conformation of amino acid residues in theory, and its main use is to evaluate the quality of the model after homologous modeling. In general, if the amino acid residues in the allowable and maximum allowable regions account for more than 90% of the whole protein, we can consider that the conformation of the model accords with the rules of stereochemistry.

Figure 1. Structures of both kinds of PETase. Surface of original PETase(A) and mutant PETase(B) both with a signal peptide head marked in orange. Ramachandran plot of mutant PETase prediction(C).

Molecular docking and comparison of mutated residues

To evaluate the affinity of the raw and mutant for the ligand (4-[(2-Hydroxyethoxy) carbonyl] benzoate, PET monomer), we performed molecular docking analysis. The binding poses and interactions of PET monomer with two proteins were obtained with Autodock Vina v.1.2.2 and binding energy for each interaction was generated (Figure 2).

The binding energy of the mutant to the ligand was -4.767 kcal/mol, while that of the original protein is -4.545 kcal/mol, indicating a higher affinity for the ligand. Comparing the binding of the ligand to the active site before and after the mutation shows that, the mutation of arginine to alanine at position 280 (marked in cyan) expands the hydrophobic active pocket and makes it easier for the polymer to bind to the active center. In addition, the mutation at position 87, 159, 238 (marked in green) changes the binding posture of the substrate to the active site and increases the affinity (one more π-bond) of the enzyme for the substrate. As reported in the literature, we speculate that the mutation at position 121, 186, 242, 246 (marked in red) changes the property of the α-helix and may increase the thermostability of the enzyme.

Figure 2. Molecular docking of enzyme and ligand. A: Original PETase; B: Mutant PETase. Amino acid residues in 5 Å is marked in pale green, ligand in sky blue, and mutant residues in red, green and cyan, respectively. The hydrogen bond is marked in yellow and π-bond in orange.

Introduce AlphaFold2

AlphaFold is an AI system developed by DeepMind that predicts a protein’s 3D structure from its amino acid sequence. It regularly achieves accuracy competitive with experiment. Unlike SWISS-MODEL, alphafold2 does not rely on homology modeling to predict protein structures, but rather on protein folding principles, which helps a lot in predicting fusion proteins.

To address the question of whether a model can determine what parts of its predictions are likely to be reliable, the researchers developed two trust measures based on the AlphaFold network.

Domain position confidence (PAE)

The predicted local distance difference test (pLDDT) score (0-100) is a per-residue confidence score, with values greater than 90 indicating high confidence, and values below 50 indicating low confidence. AlphaFold models are often shown with high confidence residues colored blue, and lower confidence in yellow, orange and red. This measure estimates whether the predicted residue has similar distances to neighboring C-alpha atoms (within 15 Å) in agreement with distances in the true structure.

3D structure prediction of fusion protein using AlphaFold2

We obtained a total of five predicted structures with similar accuracy, as can be seen in the figure 3 below. As can be seen from the PAE plot (Fig. 3A), the fusion protein has two domains, corresponding to PETase and MHETase, which are linked to each other by the active peptide. It also strongly supports the possibility of successful expression of the fusion protein, and the two enzymes do not interact with each other to greatly change their properties. Regions with scores below 50 in the pLDDT evaluation (Fig. 3B) are presumed to be signal peptide and flexible linker, respectively, which are usually amorphous structures.