1. The Ancestral Sequence Reconstruction of PETase

Recently, some researchers have suggested that profile-HMM algorithm could be used instead of BLAST local alignment algorithm for protein sequences in silico in order to circumvent low overall sequence similarity between them. Using the Pfam entry PF01738 (dienelactone hydrolase family) and the HMMER Web server to search Swiss-prot and NCBI public databases, we obtained 158 protein sequences from all species. Expected values (E-value) were between 2.4 × 10^–76 and 0.47. Lower E thresholds are more stringent, leading statistically to fewer chance-matches being reported. The 50 first bacterial protein sequences (2.4 × 10^–76 ≤ E ≤ 2.9 × 10^–6) were selected and we deleted three duplicate records. Among these protein candidates, 39 proteins have dienelactone hydrolase (DLH) domains. 21 homologous sequences were then manually added for ancestral sequence reconstruction (ASR). These sequences were aligned, and the corresponding phylogenetic tree was built using a maximum likelihood algorithm by Fireprot server (Fig. 1a).

We conducted an evolutionary analysis of PETase based on the phylogenetic tree (Fig. 1a). A lot of studies have confirmed that the certain microorganisms can secrete the extracellular enzyme to degrade biomass to support their growth and metabolism. Tracing back to the ancestral era of plastic-degrading enzymes, the α/β hydrolase superfamily differentiated into hydrolases with distinct functions and α/β structures at a certain stage (ASR3-PETase). The hallmark of α/β hydrolase folding is its remarkable plasticity, ensuring the multifunctionality of the superfamily enzymes, including lipases and esterases/cutinases, and their versatility with respect to various substrates. For example, one branch of these enzymes, the cutinases, is capable of hydrolyzing ester bonds, thus breaking down the keratin structures in plants. Nature's wonder lies in its diversity, as beyond the α/β hydrolases, certain enzymes have been endowed with peroxidase or acyl-coenzyme A (CoA) activities, such as ACOT4 HUMAN and SERINE ESTERASE, which can act on phenolic groups found in lignin or remove acetyl side chains. With the development of society, plastic became widely used as a new artificial material. Due to the structural similarity between plastics and keratin or lignin, cutinases and ligninases evolved into more effective plastic-degrading enzymes, such as PETase. Next, we sought to evaluate the PET-hydrolytic activity of reconstructed ancestral sequence (ASR1-7). As shown in Fig. 1b, ASR1-PETase and ASR7-PETase exhibited high activity in degrading PET, releasing monomer quantities of 3.44 mM and 3.82 mM, respectively. Compared with the wild type PETase, the identity of ASR1-PETase (21.37%) significantly lower than that of ASR7-PETase (67.25%) (Fig. 1c).

From the perspective of natural evolution, the evolutionary distance from ASR7-PETase to PETase was relatively close, with a limited variant space. Conversely, due to a longer evolutionary distance, obvious identity difference was observed between ASR1-PETase and PETase. For the structural analysis of PETase and ASR1-PETase, we found that both PETase and ASR1-PETase possessed the canonical α/β hydrolase domain, the conserved catalytic triad, and disulfide bonds that facilitated the stability of enzyme structure and function (Fig. 1d). Therefore, ASR1-PETase was selected as protein scaffold for further evolution through a machine learning-guided strategy.

2. A Machine Learning-guided Strategy to Engineer ASR1-PETase

During the protein application, it is necessary to improve the target protein to meet industrial conditions. The emergence of directed evolution has circumvented some restrictions to some extent, guiding proteins to evolve in a specific preset direction through random mutation and screening, but the applicability and flux limitations of directed evolution screening systems often become the bottleneck of complex protein engineering. To address this challenge, machine learning algorithm was more effective than experimental evolution. Modeling the mutation-function relationship of proteins helps to understand and target protein design, and ESM-1v is an effective machine learning model. ESM-1v can predict protein substitution even when there are few experiments or when relevant information is missing. By fine-tuning the protein family sequence model, we can find the key sites associated with mutations and efficiently predict the effects of variants on protein function. The ESM-1v model is universal and can be used to predict multiple needs without model specialization for each prediction (Fig. 2a). Comparison of ESM-1v with earlier models (such as Evmutation, DeepSequence) were shown in Table 1.

As shown in (Fig. 2c), the heat map of the replacement of sequence sites will be obtained after employing ESM-1v model, and the potential variants could be obtained according to the heat map. We select top 20 variants (including P177E, K249E, I223E, W120R, Q79N, S242G, L246H, Y46I, N81H, G201A, F213S, F275L, L122K, C287P, M17A, V265T, P177E, I223E, W120R, Q79N, S242G, L246H, Y46i, N81H, G201A, F213S, F275L, L122K, C287P. C26A, E125R, M40E, Y6A)with the highest scores, and PET-hydrolytic activity was detected through site-directed mutagenesis (Fig. 2b).

3. The High Thermostability Variants by Site-Directed Mutagenesis

In our study, we designed 20 variants to investigate the thermostability and catalytic efficiency. As shown in Fig. 3a-b, the Bis (2-Hydroxyethyl) terephthalate (BHET) conversion of both ASR1-PETase and its variants was generally higher than that of the wild type PETase. Among them, four variants ASR1-V265T, ASR1-N81H, ASR1-W120R, and ASR1-L122K exhibited increased BHET conversion at both 30 °C and 50 °C, high up to 2.71-fold of ASR1-PETase. Meanwhile, the corresponding terminal product Terephthalic acid (TPA) of four variants, was also maintained at high yields (2.58 to 2.88 mM). Furthermore, hydrolysis advantages of ASR1-V265T, ASR1-N81H, ASR1-W120R, and ASR1-L122K were also presented on PET film. As depicted in Fig. 3c, the ASR1-PETase variants maintained hydrolysis activity at 60 °C, whereas most of the wild type PETase activity was lost over 30 °C. In details, these four variants exhibited improved catalytic efficiencies on PET film at 40 °C, up to 1.58-fold (ASR1-V265T). Additionally, it was observed that the variants ASR1-V265T and ASR1-W120R maintained PET degradation products at levels of 2.40 mM and 2.69 mM, respectively, at a temperature of 60 °C, exhibiting an average increase of 2.32-fold compared to ASR1-PETase. Considering the necessity of high thermostability for industrial enzymes, we adopted a scoring strategy to identify variants with greater thermostability. We assigned different weights to the data obtained at 30 °C, 40 °C, 50 °C, and 60 °C, which were 10%, 20%, 30%, and 40%, respectively, to determine the composite scores. Based on the final scores, ASR1-V265T (Score 4.258) and ASR1-W120R (Score 3.978) was selected to analyze the product composition (Fig. 3d-f). We found that the terminal product TPA accounted for 63% to 75% of the total PET released monomers, while the remaining portion mainly comprised intermediates BHET and monohydroxyethyl terephthalate (MHET). These results indicated that enzymatic degradation of PET by a single enzyme will generate inhomogeneous products, resulting in contamination of the terminal product TPA. Considering that mixed products are unfavorable for subsequent reuse, we shifted our focus to the production of homogeneous TPA.

4. Mechanism Analysis of Increased Enzyme Activity by MD Simulations

By analyzing the stability of PETase and ASR1 variants, it could be seen that the average RMSD values of ASR1 variants were higher than PETase at 30 °C. The RMSD value of ASR1-W120R in ASR1 variants was the least, and the value of ASR1-WT was close to ASR1-V265T. This indicated that ASR1-W120R is more stable than ASR1-WT (Fig. 4a). The time-averaged total, hydrophobic, and hydrophilic SASA of PETase and ASR1 variants showed a similar trend as R_g (Fig. 4b-c). It was observed that the RMSF value of PETase fluctuated greatly at different temperatures of 30 °C and 60 °C, especially a sudden rose in the range of amino acids P155-V167, which showed that PETase had larger overall fluctuation at high temperatures and higher protein flexibility (Fig. 4d). The RMSF value of ASR1-W120R fluctuated very little at 30 ℃ and 60 ℃, indicating that the variant could sustain stability even at elevated temperatures (Fig. 4e). Through the analysis of PETase and ASR1 variants on multiple stability indexes, we found that ASR1-W120R could maintain considerable stability at high temperatures, thus, further confirmed that ASR1-W120R was an excellent PET degrading enzyme.

The solvation of the active site of an enzyme was an important factor affecting its activity. Different enzymes had different characteristics at the active site, and the binding of solvent molecules to the active site were also varied. As showed in Fig. 4f, under the conditions of 30 °C and 60 °C, the number of water molecules around the substrate binding site of ASR1 variants was more than twice as much as PETase, and the number of water molecules around the substrate binding site of ASR1-W120R and ASR1-V265T was 2 to 4 more than that in ASR1-WT. This suggested that the substrate binding site of the ASR1 variants could attract more water molecules. Meanwhile, as evident from (Fig. 4g), the active sites of the four enzymes kept getting closer to PET dimer in water as simulation time increased, and finally tended to be stable. When a stable state was reached, the distance between the active site in ASR1-WT and PET dimer was less than that between the active site in PETase and the PET dimer core. In addition, the distance between the active site and the PET dimer core of the two ASR1 variants was less than that of ASR1-WT. This suggested that compared with PETase, it was easier for the substrate to enter the active center of the ASR1 variants, therefore, promoting the transformation of the substrate. These results indicated that the activity of ASR1 variants designed by our team was better than PETase, and the activity of ASR1-W120R and ASR1-V265T variants was superior to ASR1-WT.

5. Complete TPA Conversion of PET Achieved by A Simple Two-Enzyme System

TPA produced by a single-enzyme degradation system often suffered from contamination by oligoethylene terephthalates, BHET, and MHET, which posed limitations on downstream applications. To overcome the obstacle, we have developed a two-enzyme degradation system that combines ASR1-V265T and ASR1- W120R, both performing better in preliminary validation and MD simulations, with BHETase (BsEst, identified in our previous study). It was proved that the system enables the production of sole homogeneous TPA, as illustrated in Fig. 5a. Coupling BsEst with two variants of ASR1 respectively in a two-enzyme system significantly increased TPA production compared to using variants alone in a single-enzyme system. As depicted in Fig. 5b, ASR1-V265T/BsEst and ASR1-W120R/BsEst increased TPA production by 1.7-fold and 1.8-fold over 120 h. Moreover, we observed that these two-enzyme systems produced 1.9-fold and 2.3-fold higher amounts of homogeneous TPA at 30 °C than at 60 °C.

The discovery mentioned above was further supported by the analysis of the PET film surface using scanning electron microscopy (SEM), and the measurement of reduced the water contact angle (Fig. 5c). Notably, the two-enzyme system exhibited a remarkable efficiency degrading the PET film with up to a 15.6 °C water contact angle decrease in average at 60 °C, resulting in a visibly rougher surface compared to the single-enzyme system. To implement our team's sustainability philosophy and address the additional energy costs associated with high temperatures, we also investigated the PET degradation efficiency of the two-enzyme system at room temperature (30 °C) to achieve a more environmentally friendly process. We surprisingly observed better effects at room temperature compared to 60 °C. Specifically, the average reduction in water contact angle is 23.7° at 30 °C, 1.5 fold higher than at 60 °C.

6. Homogeneous TPA Repolymerization PET to Achieve the Closed-loop PET Recycling

To achieve the closed-loop PET recycling, we designed a chemical-enzymic approach based on the homogeneous TPA through the two-enzyme system (Fig. 6a) . Initially, PET waste was degraded into the homogeneous TPA using the two-enzyme system. The homogeneous TPA was then synthesized into Dimethyl Terephthalate (DMT) under specified H₂SO₄/CH₃OH chemical conditions. Finally, DMT was directly polymerized with Ethylene Glycol (EG) and Ti(Oi-Pr)₄ to regenerate virgin PET.

Following this chemical-enzymic approach, the closed-loop recovery of PET from depolymerization to reaggregation was completed within approximately 57 h (Fig. 6a) , roughly 298 mg of PET was obtained from 654 mg of TPA. The PET was collected as a white solid after centrifugation and dried under vacuum. The crystallinity of virgin PET was detected by DSC. The heat of fusion ΔH_m and cold crystallization ΔH_c were determined by integrating areas (J g⁻¹) under peaks in DSC detection. The percent crystallinity was calculated using the following equation:

Crystallinity (%) =(ΔH_m-ΔH_c)/ΔHm×100

ΔH_m is the enthalpy of melting (J g⁻¹), ΔH_c is the enthalpy of cold crystallization (J g⁻¹), and ΔHm is the enthalpy of melting for a 100% crystalline PET sample, which is 140.1 J g⁻¹. Fig. 6b indicated that the PET crystallinity was approximately 2.30%. These results confirmed that we had successfully synthesized PET by using the homogeneous TPA obtained by the two-enzyme system, and achieved the closed-loop PET recycling.

7. Discussion

PET biodegradation represents a sustainable, low-energy solution for PET recycling, especially compared to current disposal routes such as landfill and incineration. Recent protein engineering and the more than 70 crystal structures of bacterial PET-degrading enzymes have improved our understanding of the enzyme degradation process, but there are remained challenge about the balance between high thermostability and catalytic efficiency. This project proposes to start from the ancestor sequence of PETase and deeply explore the evolution of PET, meanwhile, to provides the skeleton for the protein engineering design of PETase. Inspired by the achievements of artificial intelligence in solving the field of protein fitness to detect hidden evolutionary information, we adopted the machine learning ESM-1v model and redesigned PETase (ASR1-V265T and ASR1-W120R). Compared with wild-type ASR1-PETase, the four variants exhibited improved catalytic efficiencies on PET film at 40 °C, up to 1.58-fold (ASR1-V265T). Additionally, it was observed that the variants ASR1-V265T and ASR1-W120R maintained PET degradation products at levels of 2.40 mM and 2.69 mM, respectively, at a temperature of 60 °C, exhibiting an average increase of 2.32-fold compared to ASR1-PETase. After MD simulations, the mechanism of promoted enzyme performance has been proofed, emphasizing that the number of water molecules around the substrate binding site increased, and the distance between the substrate and the active center was closer. At the same time, the two-enzyme system constructed in this study provides homogeneous TPA for subsequent PET repolymerization, enabling the closed-loop PET recycle, and provides guidance for further investigation of other mass-produced polymer in this interesting research field.

Reference

1. Ali, S. S.; Elsamahy, T.; Al-Tohamy, R.; Zhu, D.; Mahmoud, Y. A. G.; Koutra, E.; Metwally, M. A.; Kornaros, M.; Sun, J., Plastic wastes biodegradation: Mechanisms, challenges and future prospects. Science of the Total Environment 2021, 780.

2. Guzzetti, D.; Lebrun, A.; Subileau, M.; Grousseau, E.; Dubreucq, E.; Drone, J., A Catalytically Competent Terpene Synthase Inferred Using Ancestral Sequence Reconstruction Strategy. ACS Catalysis 2016, 6 (8), 5345-5349.

3. Gricajeva, A.; Nadda, A. K.; Gudiukaite, R., Insights into polyester plastic biodegradation by carboxyl ester hydrolases. Journal of Chemical Technology and Biotechnology 2022, 97 (2), 359-380.

4. Roussel, A.; Amara, S.; Nyyssola, A.; Mateos-Diaz, E.; Blangy, S.; Kontkanen, H.; Westerholm-Pantinen, A.; Carriere, F.; Cambillau, C., A Cutinase from Trichoderma reesei with a Lid-Covered Active Site and Kinetic Properties of True Lipases. Journal of Molecular Biology 2014, 426 (22), 3757-3772.

5. Chen, C. C.; Dai, L.; Ma, L.; Guo, R. T., Enzymatic degradation of plant biomass and synthetic polymers. Nature Reviews Chemistry 2020, 4 (3), 114-126.

6. Yoshida, S.; Hiraga, K.; Takehana, T.; Taniguchi, I.; Yamaji, H.; Maeda, Y.; Toyohara, K.; Miyamoto, K.; Kimura, Y.; Oda, K., A bacterium that degrades and assimilates poly(ethylene terephthalate). Science 2016, 351 (6278), 1196-1199.

7. Fu, Y.; Wang, X.; Dong, H.; Jiang, Y. G.; Wang, M.; Xue, X.; Sigal, L., Vocabulary-Informed Zero-Shot and Open-Set Learning. Ieee Transactions on Pattern Analysis and Machine Intelligence 2020, 42 (12), 3136-3152.

8. Meier, J.; Rao, R.; Verkuil, R.; Liu, J.; Sercu, T.; Rives, A. In Language models enable zero-shot prediction of the effects of mutations on protein function, 35th Conference on Neural Information Processing Systems (NeurIPS), Electr Network, 2021.

9. Hopf, T. A.; Ingraham, J. B.; Poelwijk, F. J.; Scharfe, C. P. I.; Springer, M.; Sander, C.; Marks, D. S., Mutation effects predicted from sequence co-variation. Nature Biotechnology 2017, 35 (2), 128-135.

10. Hsu, C.; Nisonoff, H.; Fannjiang, C.; Listgarten, J., Learning protein fitness models from evolutionary and assay-labeled data. Nature Biotechnology 2022, 40 (7), 1114.

11. Hie, B. L. L.; Shanker, V. R. R.; Xu, D.; Bruun, T. U. J.; Weidenbacher, P. A. A.; Tang, S.; Wu, W.; Pak, J. E. E.; Kim, P. S. S., Efficient evolution of human antibodies from general protein language models. Nature Biotechnology 2023.

12. Shashkova, T. I. I.; Umerenkov, D.; Salnikov, M.; Strashnov, P. V. V.; Konstantinova, A. V. V.; Lebed, I.; Shcherbinin, D. N. N.; Asatryan, M. N. N.; Kardymon, O. L. L.; Ivanisenko, N. V. V., SEMA: Antigen B-cell conformational epitope prediction using deep transfer learning. Frontiers in Immunology 2022, 13.

13. Cui, H.; Vedder, M.; Schwaneberg, U.; Davari, M. D. Using molecular simulation to guide protein engineering for biocatalysis in organic solvents. Methods Mol Biol 2022, 2397, 179– 202.

14. Jerves, C.; Neves, R. P. P.; Ramos, M. J.; da Silva, S.; Fernandes, P. A., Reaction Mechanism of the PET Degrading Enzyme PETase Studied with DFT/MM Molecular Dynamics Simulations. ACS Catalysis 2021, 11 (18), 11626-11638.

15. Yin, Q.; You, S.; Zhang, J.; Qi, W.; Su, R., Enhancement of the polyethylene terephthalate and mono-(2-hydroxyethyl) terephthalate degradation activity of Ideonella sakaiensis PETase by an electrostatic interaction-based strategy. Bioresour Technology 2022, 364, 128026.