Toward Expansion of Biological Plastic Recycling: In Silico Optimization of a Polyurethane-Degrading Metagenomic Urethanase

ABSTRACT

Plastic polymer production increased to 400 million tons in 2022, yet only about 9 percent is recycled, creating growing environmental challenges. Biological plastic recycling (BPR) utilizes engineered microbial-origin enzymes to degrade synthetic polymers, but slow laboratory-directed evolution has limited BPR progress largely to poly(ethylene terephthalate) (PET). BPR research on polyurethanes (PU), a high-volume but difficult-to-recycle polymer, remains limited. This study proposes a simple, low-cost in silico enzyme-engineering workflow that combines two open-source, user-friendly generative AI tools, ProteinMPNN and AlphaFold2. We hypothesized that ProteinMPNN-guided sequence optimization, informed by AlphaFold2-predicted structure, would improve expression of a recently reported metagenomically discovered urethanase that degrades post-consumer PU. Using this workflow, the wild-type urethanase (WTU) was optimized to produce an engineered variant (AIU) that achieved a 22.6 percent increase in expression in Escherichia coli (p = 0.002). The 3D structural models of WTU and AIU were predicted with high pLDDT confidence scores. To benchmark reliability and efficiency, the workflow was also used to independently reproduce five previously reported PETase mutations, achieving near-exact laboratory-evolved mutation patterns within minutes rather than years. This proof-of-concept study is among the first to demonstrate in silico urethanase sequence optimization and to validate ProteinMPNN and AlphaFold2 as useful tools in expanding BPR research to polyurethane degradation.

INTRODUCTION.

Plastics exist in many polymer forms and are widely used in modern life. Nearly 400 million tons are produced annually, yet only about 9 percent is recycled. The remainder accumulates in landfills, incineration sites, and natural environments where it poses environmental and human health risks [1–3]. Biological plastic recycling (BPR) offers a more sustainable alternative but has remained minimally exploited beyond poly(ethylene terephthalate) (PET) [3]. Plastic-degrading enzymes arise rarely in nature and once discovered are usually improved through directed laboratory evolution, which is resource-, skill-, and time-intensive [4]. Although metagenomic approaches now enable the discovery of natural plastic-degrading enzymes, engineered improvement remains necessary because most natural enzymes lack sufficient efficiency for industrial applications [3,5]. Since the identification of Ideonella sakaiensis PETase (IsPETase) in 2016, many successful engineering efforts have been reported, yet similar progress for other polymers has lagged due to the slow pace of natural enzyme discovery as well as enzyme optimization [3,6–8].

Polyurethane (PU) recycling rates are among the lowest of high-demand polymers despite PU being widely used in insulation, automotive components, bedding, furniture, coatings, footwear, and apparel [3,5,9]. A true urethanase capable of completely cleaving urethane bonds into alcohol, carbon dioxide, and aromatic diamine was only recently identified by metagenomic analysis of soil from a PU waste site in Germany [5]. In that study, a two-step chemoenzymatic process enabled partial circular PU reuse, but the required high-temperature glycolysis step was hazardous and generated toxic intermediates [5]. Because the urethanase was discovered from environmental DNA, no native host organism is known, and no UniProt ID or solved structure exists at the time of this research and manuscript preparation. Among the three enzymes reported by Branson et al., this study focuses on UMG-SP1, here referred to as wild-type urethanase (WTU), which exhibits limited expression and usability in standard laboratory expression systems [5].

Recent advances in protein engineering have reduced uncertainties and time requirements in enzyme redesign across biotechnology and pharmaceutical applications. Computational redesign tools such as PROSS (Protein Repair One-Stop Shop) apply physics-based energy calculations to improve protein stability and expression [10]. Deep learning approaches such as ProteinMPNN (Protein Message Passing Neural Network) generate optimized amino acid sequences based on a supplied 3D backbone [11–12]. AlphaFold2 enables accurate structural prediction even when crystallographic structures are unavailable [13], making it possible to engineer metagenomically discovered enzymes without prior structural data. The developers of ProteinMPNN and AlphaFold2 shared the 2024 Nobel Prize in Chemistry, underscoring the robustness and transformative potential of these approaches.

Because metagenomically discovered enzymes may lack sequence features compatible with efficient expression in standard laboratory hosts [14], we hypothesized that ProteinMPNN-guided sequence optimization, using an AlphaFold2-predicted backbone as structural input, would increase the expression yield of WTU in Escherichia coli. To evaluate the reliability of this approach, we also compared the workflow’s mutation outputs to previously reported PETase laboratory-evolved mutation patterns [4,7]. This study therefore tests whether in silico sequence optimization can improve WTU expression and whether the workflow can reproduce known enzyme engineering outcomes.

MATERIALS AND METHODS.

Selection and Use of AI Tools.

All computational work was performed on a standard internet-connected laptop. The goal was to establish a workflow using freely accessible AI tools that do not require programming expertise. AlphaFold2 and ProteinMPNN were selected for structure prediction and sequence optimization due to their high accuracy and availability on web-based platforms such as ColabFold and ColabDesign [15,16]. These user-friendly interfaces allow users to submit a sequence or structure file and receive results without command-line processing.

Workflow Overview.

The workflow consisted of four main steps:

Step 1. Structure prediction: The metagenomic urethanase sequence reported by Branson et al., 2023 [5] was submitted to AlphaFold2 to obtain a predicted 3D structure for use in subsequent redesign.

Step 2. Sequence optimization: The AlphaFold2-generated structure coordinate file was submitted to ProteinMPNN to generate redesigned protein sequences. If a Protein Data Bank structure were available at the time of this work, this step could be performed directly without Step 1.

Step 3. DNA sequence generation: Redesigned protein sequences were converted to codon-optimized DNA sequences for Escherichia coli expression using DNAWorks [17].

Step 4. Structural verification: Redesigned variants were resubmitted to AlphaFold2 to assess folding and predicted Local Distance Difference Test (pLDDT) scores. pLDDT ≥70 was considered acceptable for experimental testing and ≥90 high confidence.

Workflow Validation Using PETase.

To evaluate whether ProteinMPNN could reproduce experimentally derived enzyme variants, the native PETase structure (Protein Data Bank ID: 5XJH) was submitted to ProteinMPNN. The server returned 10 redesigned sequences. Five PETase amino acid replacements reported from laboratory evolution studies [4,7] were compared to the corresponding positions in the redesigned sequences. A two-tailed z-test of proportions (n₁ = 10, n₂ = 10) was used to determine whether observed replacement frequencies differed significantly from reported laboratory-evolved frequencies.

Structure Prediction and Design of WTU, AIU, and PRU.

No urethanase structure was available in the Protein Data Bank or AlphaFold2 Protein Structure Database due to the absence of a UniProt ID at the time of this work in 2024. The urethanase sequence from Branson et al. [5] was submitted to AlphaFold2 and the top-ranked structure was selected as WTU. ProteinMPNN generated 15 redesigned sequences of this WTU, and the first was selected as AIU (AI urethanase). For comparison, PROSS was also used and returned eight redesigned variants; the top-scoring one was designated PRU (PROSS urethanase).

DNA Synthesis, Cloning, and Transformation.

DNAWorks was used to generate codon-optimized DNA sequences for WTU, AIU, and PRU. Gene synthesis and cloning into pET28A were performed by GenScript (Piscataway, NJ). Plasmids were transformed into E. coli BL21 cells by heat shock and plated on LB agar containing kanamycin.

Protein Expression and SDS-PAGE Analysis.

Single colonies were grown in LB-kanamycin and induced with IPTG. Cells were lysed in Laemmli buffer and supernatant fractions were separated by SDS-PAGE. Gels were stained with Coomassie blue. Experiments were performed in triplicate. Relative band intensities were compared by measuring band thickness using a metric ruler, a widely used semi-quantitative approximation when densitometry is not available. Statistical significance of differences in expression levels between variants was evaluated using a two-tailed Student’s t-test (n = 3).

RESULTS.

WTU Structure Prediction.

AlphaFold2 generated five predicted structures for WTU; the highest-ranked model was selected. Approximately 99 percent of residues displayed pLDDT scores greater than 90, indicating high structural confidence, with only 3 residues scoring below 70 (Figure 1).

Sequence Redesign Using ProteinMPNN and PROSS.

PROSS returned nine redesigned WTU variants and recommended one containing approximately 16 percent amino acid substitutions. ProteinMPNN generated 15 redesigned variants, each containing more than 40 percent substitutions. Multiple sequence alignment comparisons of WTU, AIU, and PRU were performed using MUSCLE (Figure S1).

Structural Confidence of Redesigned Variants.

AIU and PRU were resubmitted to AlphaFold2 for structural verification. AIU displayed pLDDT greater than 90 for approximately 99 percent of residues (Figure 2), indicating a confidently folded structure suitable for expression testing. The pLDDT score was similarly high for PRU.

Protein Expression in E. coli.

SDS-PAGE analysis showed that AIU expression in E. coli was 22.6 percent higher than WTU (p = 0.0019, Student’s t-test) (Figure 3). PRU expression was visibly lower than WTU and was not quantified.

These results support the hypothesis that ProteinMPNN-guided sequence optimization can increase heterologous expression of a metagenomic urethanase.

Workflow Validation Using PETase.

ProteinMPNN reproduced each of the five laboratory-evolved PETase amino acid replacements in 9 of 10 or 10 of 10 redesigned sequences (Figure 4). Two-tailed z-tests yielded p ≥ 0.3 for all positions, indicating no statistically significant difference from the laboratory-evolved replacement pattern. This redesign process was completed in minutes, representing approximately over a thousand-fold time savings compared to laboratory evolution process that can take several months or more to achieve.

DISCUSSION.

Biological plastic recycling (BPR) is gaining research attention, yet meaningful progress has been concentrated largely on PET, while comparable advances for other polymers have remained limited [3]. Expanding BPR to additional polymer classes, including PU, is important due to their high production volume and limited recycling options [9]. In this context, this study demonstrates a proof-of-concept workflow for sequence optimization and heterologous expression of a metagenomically discovered urethanase recently reported to depolymerize PU intermediates [5]. The urethanase, which lacked a solved or predicted structure in public databases, also lacked documented expression adaptation for use in standard enzyme production systems.

This work is among the first to apply the combined use of ProteinMPNN and AlphaFold2 in BPR enzyme research. AlphaFold2 enabled structural prediction of the WTU, which then allowed ProteinMPNN-guided sequence optimization to generate an engineered variant: AIU. The optimized AIU enzyme demonstrated a 22.6 percent increase in expression in Escherichia coli compared to WTU, supporting the hypothesis that in silico sequence optimization using ProteinMPNN can improve expression compatibility of metagenomic enzymes in heterologous systems. To evaluate workflow reliability and efficiency, the same process was used to independently reproduce five previously reported PETase laboratory-evolved mutations, achieving near-exact mutation patterns within minutes rather than the multi-year experimental evolution timelines reported in earlier PETase work [4,7].

The workflow demonstrated here is simple, rapid, and cost-effective, as both ProteinMPNN and AlphaFold2 are openly accessible and require no specialized computational expertise. Future work will be necessary to characterize AIU’s catalytic activity, stability, substrate specificity, and performance under conditions relevant to PU depolymerization. Such experiments require specialized facilities capable of handling hazardous glycolysis conditions and PU intermediate substrates [5].

Overall, this proof-of-concept study indicates that generative AI-guided sequence optimization can accelerate early-stage BPR enzyme development and may help expand BPR research to polymer classes beyond PET. Further biochemical validation is required to determine whether increased expression translates to useful catalytic function for polyurethane degradation. Increased heterologous expression can improve the practical utility of metagenomic enzymes by enabling higher enzyme yield which is a prerequisite for downstream biochemical characterization and application including testing for catalytic activity. Metagenomic enzyme discovery and engineering have shown that expression optimization is often a critical first step toward enabling functional validation.

ACKNOWLEDGMENTS.

This work was self-funded. We thank GenScript for gene synthesis services. Computational work used open-source web tools.

SUPPORTING INFORMATION.

Supporting information includes:

Figure S1.  Sequence alignment of engineered urethanase variants.

REFERENCES.

1. Geyer, J.R. Jambeck, K.L. Law, Production, use, and fate of all plastics ever made. Sci. Adv. 3, e1700782 (2017).
2. Plastics Europe, Plastics – the Fast Facts 2023 (PlasticsEurope, Brussels, 2023).
3. J. Acosta, H.S. Alper, Advances in enzymatic and organismal technologies for the recycling and upcycling of petroleum-derived plastic waste. Curr. Opin. Biotechnol. 84, 103021 (2023).
4. Lu et al. Machine learning aided engineering of hydrolases for PET depolymerization. Nature 604, 662-667 (2022).
5. Branson et al. Urethanases for the enzymatic hydrolysis of low molecular weight carbamates and the recycling of polyurethanes. Angew. Chem. Int. Ed. 62, e202216220 (2023).
6. Yoshida et al. A bacterium that degrades and assimilates poly(ethylene terephthalate). Science 351, 1196-1199 (2016).
7. Z.L. Zhong-Johnson et al. Analysis of poly(ethylene terephthalate) degradation kinetics of evolved IsPETase variants using a surface crowding model. J. Biol. Chem. 300, 105783 (2024).
8. OECD, Global Plastics Outlook (OECD Publishing, Paris, 2022).
9. Liu et al. Biodegradation and upcycling of polyurethanes: Progress, challenges, and prospects. Biotechnol. Adv. 48, 107730 (2021).
10. J. Weinstein, A. Goldenzweig, S.Y. Hoch, S.J. Fleishman, PROSS 2: A new server for the design of stable and highly expressed protein variants. Bioinformatics 37, 123-125 (2021).
11. Dauparas et al. Robust deep learning based protein sequence design using ProteinMPNN. Science 378, 49-56 (2022).
12. H. Sumida et al. Improving protein expression, stability, and function with ProteinMPNN. J. Am. Chem. Soc. 146, 2054-2061 (2024).
13. Jumper et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583-589 (2021).
14. -Y. Colin et al. Ultrahigh-throughput discovery of promiscuous enzymes by picodroplet functional metagenomics. Nat. Commun. 6, 10008 (2015).
15. Mirdita et al. ColabFold: Making protein folding accessible to all. Nat. Methods 19, 679-682 (2022).
16. Varadi et al. AlphaFold Protein Structure Database: Massively expanding structural coverage of protein space. Nucleic Acids Res. 50, D439-D444 (2022).
17. M. Hoover, J. Lubkowski, DNAWorks: An automated method for designing oligonucleotides for gene synthesis. Nucleic Acids Res. 30, e43 (2002).

---

---

Posted by buchanle on Thursday, May 14, 2026 in May 2026.

Tags: AlphaFold2, biological plastic recycling, ProteinMPNN, Urethanase engineering