Earth’s ecosystem is approaching a planetary-scale transformation as a result of human influence1. A wide variety of petroleum-based synthetic polymers are produced worldwide, with a 20-fold increase in annual production since the 1960s, and a slowdown of this trend is not expected2. Marine pollution and the negative impacts of microplastic exposure on human health have caused escalating public and governmental concerns, boosting the demand for a circular plastic economy3-7. In this regard, biotechnological plastic recycling has become a thriving research area in recent years8-10. Poly(ethylene terephthalate) (PET) is one of the most widely used man-made synthetic plastics worldwide, with an annual manufacturing capacity of over 30 million tons2. Over the past two decades, research on the biodegradation of PET has progressed from detection of trace amounts of released products to highly efficient degradation11-12. A breakthrough was achieved in 2020 with the development of an engineered LCC variant (LCCICCG) that exhibited 90% depolymerization of pretreated PET waste at high PET concentrations (200 g kg-1). Further synthesis of virgin polymers by using the released monomers offered a glimpse into the application of biocatalysis in plastic recycling at the industrial level13. However, the 10% conversion loss due to the nonbiodegradable residual PET waste with a large increase in crystallinity represents an unmet challenge (Fig. 1A). Such limitations can be vastly suppressed by decreasing the reaction temperature, but the catalytic efficiency of PET hydrolases is concomitantly sacrificed14, which is not desirable as the reaction duration has a greater impact on the process cost than the energy cost of maintaining elevated operating temperatures12, 15. In most homogeneous reactions, high enzyme loading is generally preferred for maximizing reactor productivity. However, as a typical surface erosion process, enzymatic hydrolysis of PET can hardly permeate the inner core of the polymer, resulting in a limited number of superficial ester bonds (also termed attack sites) being accessed even if the enzymes are in great excess16, 17. Saturation thus occurs when all attack sites on the surface become occupied, and the excess enzyme molecules accumulate in the solvent (Fig. 1B). Therefore, the desired improvement in depolymerization efficiency can hardly be accomplished by simple process optimization, increasing the demand for new functional enzymes with balanced thermostability and high hydrolytic efficiency.
Previous efforts have been expended in the quest for new PET hydrolases as well as in the optimization of known enzymes18-22. A recently reported variant of IsPETase using a machine-learning-aided approach (FastPETase) affords improved depolymerization efficiency compared to that of other IsPETase mutants at 50 °C18. More recently, directed evolution was used to generate HotPETase with a Tm of 82.5 °C, which degrades PET more efficiently than LCCICCG after 1 h of reaction at low enzyme loading (0.29 mgenzyme g-1 enzyme loading) at 65 °C19. However, in heterogeneous catalysis reactions, different enzymes may exhibit dramatically distinct catalytic efficiencies at low and high enzyme concentrations, as in the case of cellulases16. Erickson et al. also demonstrated that IsPETase showed similar hydrolytic conversion compared to its variant IsPETaseW159H/S238F at low enzyme loading (0.5-1 mgenzyme gPET-1), but approximately 2-fold lower conversion at higher enzyme loading (3 mgenzyme gPET-1)23. Since different studies employ varying experimental conditions, the designed PET hydrolases may not maintain their reported catalytic efficiency at the industrial level. Therefore, we evaluated the depolymerization performance of LCC, LCCICCG, BhrPETase, FastPETase, and HotPETase at various enzyme loading levels, but unfortunately, at enzyme saturation conditions, none of the engineered PET hydrolases tested exhibited higher catalytic performance than LCCICCG except for BhrPETase, which showed similar hydrolysis efficiency (Fig. 1C).
The last few years have witnessed impressive progress in the tailoring of natural enzymes by computational redesign strategies24. Given our limited knowledge of how a sequence encodes catalytic function in polymer-degrading enzymes, exploiting physics-based computational redesign and rational design approaches is suboptimal for improving the interfacial reaction efficiency. Inspired by the achievements in artificial intelligence for addressing the protein fitness landscape to probe hidden evolutionary information, we employed a computational strategy that incorporates a protein language model and force-field-based engineering algorithms to address the aforementioned challenge. The redesigned PET hydrolase (TurboPETase) derived from this campaign exhibited a 4.4-fold improvement in PET-specific activity compared to that of wild-type BhrPETase and LCCICCG under enzyme saturation conditions. The extraordinary degradation performance afforded by TurboPETase allowed nearly 100% depolymerization toward untreated PET containers in merely one day, as well as the full degradation of pretreated postconsumer PET bottles and lower-grade PET products at an industrially relevant scale, addressing the challenge with regard to the residual nonbiodegradable PET waste and rendering this highly efficient, optimized enzyme a good candidate for future applications in industrial plastic recycling processes. The mechanism underlying the promotion of enzyme performance has been demonstrated via kinetic analyses derived from an inverse Michaelis‒Menten reaction regime as well as structural analysis, highlighting the importance of the ability to target specific attack sites on the polymer surface.
Computational redesign of an efficient PET hydrolase.
Although nature has evolved a family of hydrolases to degrade PET polymers, they share common structural features (a canonical catalytic triad and oxyanion hole) with other esterases that act on small molecules and comparable conversion rates for soluble esters12. A quantum mechanics/molecular mechanics (QM/MM) study revealed that the free energy barrier and hydrolytic reaction thermodynamics do not differ significantly between IsPETase and the LCCICCG mutant25. Therefore, it is likely that the ability to attach to the PET surface, rather than the core catalytic motif, has been modified during the evolutionary trajectory of hydrolases to enhance the depolymerization of insoluble plastics, as with other natural polymer-degrading enzymes. Describe the mapping from protein sequence to function through deep learning-based models has proven to be successful in many cases26-28. Divergence times that indicate the improvement in polymer degradation during evolution may be captured from the relative fitness of variants of a protein along the evolutionary landscape.
To this end, we employed a language model trained on two datasets that involved approximately 26,000 homologous sequences of PET hydrolases and other esterases across evolution, to predict the probability of amino acid variation from the evolutionary landscape (Fig. 2A). BhrPETase and LCCICCG were used as the inputs due to their relatively high thermostability and hydrolytic activity. A Transformer encoder was used to process input amino acid sequences with absolute position embedding. Residue positions are sorted by the mean of the logits of 19 mutations assigned to the wild type amino acid at each position. The top ten amino acid positions with the highest average scores of each model were selected. After removing duplicated amino acid positions from the 40 top candidates, 18 amino acid positions where the wild-type residues fit less well than potential substitutions were obtained (Supplementary Table S1).
According to the crystal structures of BhrPETase (PDB code: 7EOA) and LCCICCG S165A in complex with MHET (PDB code: 7VVE29), 7 of the 18 generated amino acid positions (W104, H164, M166, W190, H191, H218, and F/I243) were suggested to be embedded in a PET-binding groove (Fig. 2B). Recently, Chen et al. reported that the Ser214/Ile218 double mutants of several IsPETase-homologous enzymes had enhanced PET hydrolysis activity (by at least 1.3-fold) but vastly decreased Tm values (by approximately 10 °C)30. Since His218 (corresponding to Ser214 in IsPETase) was involved in our predicted candidates, we initially explored the hydrolytic performance of the H218S/F222I variant on both BhrPETase and LCCICCG. As expected, BhrPETaseH218S/F222I and LCCICCG/H218S/F222I variants demonstrated a 2.6-fold and 1.9-fold increase in PET-hydrolytic activity, respectively, compared to the wild-type scaffolds in the hydrolysis of Gf-PET films at 65 °C under enzyme saturation conditions (Supplementary Table S2). In particular, the BhrPETaseH218S/F222I variant yielded 1.4-fold greater levels of degradation products than the LCCICCG/H218S/F222I variant, suggesting that this engineered enzyme is a more suitable candidate for further engineering. Using the BhrPETaseH218S/F222I (referred to as BhrPETase M2) variant as the starting point, we subjected the remaining 6 positions (W104, H164, M166, W190, H191 and F243) to generate 32 variants (Supplementary Table S3). After experimental verification, W104L, W104S, W104H, W104G, F243I, F243T, and F243G resulted in improved hydrolytic activities (by 10% to 34%).
Notably, the BhrPETase M2 variant exhibited a melting temperature of 85 °C, which was 11 °C lower than that of the wild-type enzyme. The active mutations at the W104 and F243 positions on BhrPETase M2 reduced the stability even further, with Tm values ranging from 71.5 °C to 84 °C. Empirical data have suggested that an at least 12 °C higher Tm than the Topt is needed to enable catalyst longevity20. Hence, a PET hydrolase with a Tm over 77-85 °C is preferred for efficient PET degradation. The nonnegligible decrease in the stability of the active variants limited further combination, and compensatory mutations needed to be introduced first to suppress the deleterious effects. Therefore, we applied our previously devised GRAPE strategy21, which employs four complementary algorithms, namely, FoldX31 (force field energy function), Rosetta_cartesian_ddg32 (force field energy function), ABACUS33 (statistical energy function) and DDD34 (force field energy function), to design stabilizing mutations to compensate and buffer the destabilizing mutations (Fig. 2C). Upon experimental validation, 3 beneficial variants (A209R, D238K, and A251C-A281C) resulted in improved thermostability without compromising the activity (Supplementary Table S4). We added the stabilizing variants to BhrPETase M2 using a stepwise combining strategy and resulted in a BhrPETase M6 variant (BhrPETaseH218S/F222I/A209R/D238K/A251C/A281C), which exhibited a restored melting temperature of 97 °C without sacrificing activity. Subsequently, active mutations at the W104 and F243 positions were combinatorically assembled and accumulated onto the thermostable BhrPETase M6 variant, generating 12 new variants (Fig. 2D). After comparative analysis of specific activities and melting temperatures (Supplementary Table S5), the best combination variant, BhrPETaseH218S/F222I/A209R/ D238K/A251C/A281C/W104L/F243T (referred to as TurboPETase), was selected with a Tm of 84 °C and a significant 4.4-fold improvement in PET-specific activity towards GF-PET films compared to wild-type BhrPETase.
We subsequently evaluated the depolymerization rate of TurboPETase with respect to other PET hydrolases under enzyme saturation conditions across a range of temperatures from 40 to 72 °C (Supplementary Fig. S1). The degradation results highlight the substantially superior hydrolysis activity performance of TurboPETase at all the temperatures tested, especially at elevated temperatures of 65 °C and 72 °C (Fig. 2E). At 65 °C, BhrPETase, LCCICCG, HotPETae and FastPETase achieved 5.2%, 5.2%, 2.6%, and 0.6% depolymerization within 3 h, a time course over which the reaction progressed linearly. At the same temperature, TurboPETase afforded 21% depolymerization, which was 4-, 4-, 8- and 33-times higher than those of BhrPETase, LCCICCG, HotPETae and FastPETase, respectively. At 72 °C, the depolymerization of TurboPETase was improved to 32% in 3 h, whereas other PET hydrolases reached 0.3%-9% conversion. Even at 50 °C, TurboPETase can achieve 3% depolymerization, which is 1.8-fold higher than that of FastPETase after 3-h reaction. It is worth noting that the higher depolymerization performance was mainly attributed to the specific hydrolysis activity rather than the greater product accumulation during long-term survival due to the enhanced longevity. This is not to say that TurboPETase can not suffer from prolonged periods of elevated temperatures. At 65 °C, extending the incubation time to 24 h accomplished approximately 100% depolymerization, which confirmed the thermotolerance and catalytic performance of TurboPETase.
The successful design encouraged further exploration of the generalizability of the predicted active mutations in other PET hydrolases. The W104L and F243T mutations were introduced to several homologous enzymes, including LCCICCG, Tfh from Thermobifida fusca35, and Tfcut2 from T. fusca KW336(Supplementary Fig. S2). To our delight, the resulting W104L variants of LCCICCG, Tfh (in which the corresponding residue is W109L) and Tfcut2 (in which the corresponding residue is W109L), showed enhanced hydrolytic activity on Gf-PET films (up to a 6.9-fold increase), albeit with reduced thermostabilities relative to those of their respective scaffolds (Tm decreased by 4-6 °C), thus showcasing the portability of this mutation to other PET hydrolases. In contrast to the W104L mutation, the F243T variants of Tfh (in which the corresponding residue is F249T) and Tfcut2 (in which the corresponding residue is F249T) showed no beneficial effect on enzyme activity, suggesting that this mutation may only affect the hydrolytic activity of BhrPETase and its highly homologous proteins.
Mechanism underlying the improved properties of TurboPETase via an inverse Michaelis‒Menten approach and structural analysis
The depolymerization performance under enzyme saturation conditions can provide important guidance for further efforts in engineering enzymes for industrial promise. However, in the absence of physically meaningful kinetic parameters, one can typically perform superficial analyses to interpret the molecular mechanisms of the catalytic process. As this is a heterogeneous process, the real accessible molar concentration of the substrate is unknown, which makes the use of textbooks on enzyme kinetics challenging. An inverse Michaelis‒Menten equation has been successfully employed to study the kinetics of heterogeneous enzymes such as cellulases, by which the catalytic efficacy against accessible attack sites on the PET surface can be estimated37-39.
In this study, the kinetic parameters of both the conventional Michaelis‒Menten model and inverse Michaelis‒Menten model were determined by fitting for several PET hydrolases at 65 °C on Gf-PET films. The conventional Michaelis‒Menten model examines reaction regimes in which the substrate is present at substantially higher levels than the biocatalyst, whereas the inverse Michaelis‒Menten model studies the reaction with enzyme in excess of the substrate. As shown in Fig. 3A, the curves showed near-linear relationships of the initial rate and substrate load under all conditions for TurboPETase, BhrPETase and LCCICCG. Conventional saturation behaviour was not observed because even the lowest enzyme concentrations used here (0.12 μM) were too high for the conventional approach to be valid, which indicates the very fast rates of dissociation of the enzymes from the PET surface. Therefore, increasing the substrate concentration seems to have a marginal effect on conversion yields at the same enzyme/substrate ratio. Further research with more sophisticated techniques is needed to determine the molecular origins of the high convKM and convVmax values for these PET hydrolases.
Although these enzymes could not meet the criteria for the conventional approach, the inverse approach was more applicable. The kinetic parameters invKM and invVmax/massS0, derived from the nonlinear regression analyses in Fig. 3B are listed in Table 1. The results for invKM in Table 1 revealed marginal differences between TurboPETase, BhrPETase, and LCCICCG, especially for the values at 30 g L-1 substrate loading, indicating that the adsorption capacity of TurboPETase at the attack sites on the PET surface was not impaired. It’s noteworthy that TurboPETase exhibited a 3.3-fold increase in maximal reaction velocity per available reactive site (invVmax/S0) compared with BhrPETase and LCCICCG. Since no substantial differences in invKM values were observed between TurboPETase and its parental enzymes BhrPETase and LCCICCG, we presumed that this behaviour should rely, at least in part, on the ability to attack a large subset of specific attack sites that can be hydrolysed to form a productive complex.
Table 1. Kinetic parameters of TurboPETase, BhrPETase, and LCCICCG derived from the inverse Michaelis‒Menten experiments.
Parameters
|
invKM (mgenzyme gPET-1)
|
invVmax/S (μmol g-1 s-1)
|
|
Substrate loading
|
|
12 g L-1
|
20 g L-1
|
30 g L-1
|
12 g L-1
|
20 g L-1
|
30 g L-1
|
TurboPETase
|
0.40±0.11
|
0.33±0.05
|
0.16±0.02
|
0.087±0.004
|
0.094±0.002
|
0.078±0.001
|
BhrPETase
|
0.52±0.19
|
0.26±0.11
|
0.18±0.03
|
0.026±0.002
|
0.028±0.002
|
0.025±0.001
|
LCCICCG
|
0.35±0.03
|
0.28±0.12
|
0.16±0.02
|
0.023±0.001
|
0.026±0.001
|
0.024±0.001
|
According to the model of TurboPETase predicted by AlphaFold2 and subsequent molecular dynamics (MD) simulations, key aspects of the improved performance of TurboPETase were proposed as follows: improved flexibility of the substrate binding cleft (H218S/F222I, W104L and F243T), optimized charge‒charge interactions at the protein surface (A209R and D238K), and introduction of a disulfide bond (A251C-A281C). A209R, D238K, and the disulfide bond A251C-A281C are suggested to primarily contribute to improving thermostability while maintaining activity (Supplementary Fig. S3), whereas the enhanced hydrolysis efficacy may be attributed to substitutions in proximity to the active sites. Chen et al. found that PET hydrolytic activity could benefit from the higher flexibility of the active site in the H214S/F218F double mutant 30. This trend was maintained by the synergistic interactions conferred by the addition of W104L and F243T, as revealed by the Cα root-mean-square fluctuation (RMSF) results (Fig. 3C). The greatly increased flexibility along the PET-binding groove was suggested to provide more space to accommodate a variety of attack conformations through dynamic binding, which may be crucial for the formation of catalytically competent complexes on different surface structures (Fig. 3D). Based on the above results, we reasoned that TurboPETase is probably more promiscuous with respect to the conformation of the PET strand it attacks. Nevertheless, detailed analysis of the mechanism requires further efforts through more in-depth research.
Complete degradation of postconsumer PET products at a large scale
Beyond the model substrates, we collected and determined the crystallinity of 19 raw, untreated postconsumer PET products used in the packaging of food, beverages, office supplies, and household goods available at local grocery store chains (Supplementary Fig. S4). Previous studies indicated that PET containers with less than 10% crystallinity could be fully degraded by PET hydrolases in 1 week, whereas a high degree of crystallinity above 20% of PET bottles and fibres led to a considerably low enzymatic degradation rate14, 18. In the present study, all PET samples with less than 15% crystallinity from 10 untreated postconsumer PET food containers were fully degraded by TurboPETase in merely one day (Fig. 4A and Supplementary Fig. S5). Depolymerization of a complete untreated postconsumer PET container (roughly 5.5 g with 10.5 cm length, 10.5 cm width, 3.5 cm height) was further performed by TurboPETase at 2 mgenzyme gPET-1 at 65 °C. The time-course depolymerization analysis revealed an almost linear rate during the first 12 h, and a subsequent linear decay rate was observed, with full degradation accomplished in 18 h (Fig. 4B).
Despite the PET containers, which can be biodegraded in situ without any treatment, we evaluated the performance of TurboPETase on 9 untreated PET products with crystallinity above 20%. Unfortunately, a maximum depolymerization of 4% was reached within 24 h, with the levels of subsequent degradation products not increasing over time (Supplementary Fig. S6). The addition of fresh enzymes did not give rise to any additional products, suggesting that the reaction was not halted by catalyst deactivation. These results were consistent with a previous suggestion that crystallinity might be the most important factor that influences enzymatic degradation20. Therefore, for industrial-scale biocatalytic recycling of highly crystalline PET products, pretreatment by thermomechanical amorphization is essential for providing uniformly degradable substrates. We further explored the scale-up depolymerization of 200 g L-1 and 300 g L-1 postconsumer coloured PET bottle wastes to evaluate the maximal reactor productivity. A melt-quenching pretreatment of the highly crystalline PET materials was applied accordingly to amorphize and increase the number of surface attack sites. To our delight, nearly complete depolymerization of postconsumer coloured PET bottle wastes was achieved in 8 h and 10 h for substrate loading levels of 200 g L-1 and 300 g L-1 at 2 mgenzyme gPET-1 at 65 °C, respectively (Fig. 4C). The results demonstrate a significantly higher depolymerization performance of TurboPETase at 65 °C relative to LCCICCG at elevated temperatures, and highlight the suppression of the “physical aging” process to eliminate the 10% conversion loss. In addition to the achieved full degradation with almost no residual PET waste, the successful implementation at 300 g L-1 substrate loading in the present study rendered a scale-up production feasible, which may impact an approximately 20% decrease in the minimum selling price (MSP) of recycled TPA (from $2.25/kg to $1.78/kg), whereas the total enzyme cost only accounts for 6% of the overall MSP15. The initial rate achieved a maximum productivity of 77.3 gTPAeq. L-1 h-1, which was 1.8-fold higher than the productivity reported previously by using LCCICCG with postconsumer coloured-flake PET waste13. Degradation at higher substrate loading levels was also evaluated, exhibiting unsacrificed enzymatic performance but compromised hydrolysis efficiency due to the precipitation of supersaturated TPA monomers.
Encouraged by the superior degradation performance of TurboPETase, we subsequently attempted to deconstruct lower-grade PET products made from recycled PET bottle flakes. Large scale depolymerization of coloured strappings at 300 g L-1 substrate loading also led to ~100% conversion in 16 h, with a maximum productivity of 66.3 gTPAeq. L-1 h-1 (Fig. 4D). The PET strapping powders after sieving were more inhomogeneous than the PET bottle powders due to their higher hardness. Since the particle size can influence the biocatalytic degradability of PET, it may partly explain the hysteresis of the depolymerization of PET strapping. Although recycled PET recovered from bottles can be used in other lower-grade PET applications, postconsumer lower-grade products can no longer be recycled via mechanical recycling and will thus ultimately find its way to accumulate in the environment40. Complete degradation of the lower-grade PET products with TurboPETase at a high industrially relevant scale provides a potential route for full circularity of PET materials.