Polyurethane is not just another plastic waste problem. It is a chemistry problem with its own bonds, its own structure, and its own recycling obstacles.
A research team has now designed an enzyme that can help break polyurethane down into the basic chemicals used to make it. The work shows how neural networks can move enzyme discovery beyond searching nature for an existing solution and toward designing proteins for a specific industrial challenge.
Why polyurethane is difficult to recycle
Plastic pollution is often described as one crisis, but the source article makes clear that it is better understood as many related problems. Different plastics are built from different polymers, and those polymers are held together by different chemical bonds. A process that works on one polymer may not work on another.
That is why earlier progress with enzymes that act on common plastics like polyesters and PET only solves part of the wider issue. Polyurethane has a different chemistry, and it is widely used in materials such as foam cushioning.
The scale is large. According to the paper described in the source, 22 million metric tons of polyurethane were made in 2024.
The defining urethane bond includes a nitrogen attached to a carbon, which is also attached to two oxygens. One of those oxygens links into the rest of the polymer. The surrounding polymer structure can be complex, and it often includes ringed structures related to benzene.
That complexity matters. Polyurethane chains are often heavily cross-linked, and bulky structures can block enzymes from reaching the bonds they need to digest. In plain terms, the target is not sitting out in the open.
The limits of the existing chemical route
The researchers did not begin by trying to replace the whole recycling process. Instead, they looked for an enzyme that could fit into a process already involving diethylene glycol.
Diethylene glycol can partly break down polyurethane molecules, but it requires elevated temperatures. It also leaves behind a complicated mixture of chemicals that cannot be put back into useful reactions. The source article says this material is typically incinerated as hazardous waste.
That makes the goal clear: a better process would not only attack polyurethane, but would also steer the reaction toward the basic building blocks that can be used again.
The team first tested the enzymes already reported in the literature as capable of breaking down polyurethanes. Out of 15 enzymes, only three showed decent activity against the polymer being tested. Even those mostly failed to reduce the polymer to its constituent starting materials.
The best of those existing enzymes became the starting point for a broader search. Researchers looked for related proteins in public databases and used the AlphaFold database of predicted structures to identify more distant proteins with similar folds. These candidates were not strong solutions on their own, but they supplied useful information for training an AI system.
How the neural network narrowed the search
The team used Pythia-Pocket, a neural network focused on identifying whether amino acids in a protein are likely to contact chemicals that the structure can bind, along with other functional features. They combined it with Pythia, another neural network that predicts whether a protein is likely to form a stable structure.
The researchers were not simply looking for a protein with the right overall shape. They wanted a candidate with a practical balance: ordered enough to form a useful binding pocket, but flexible enough to work around different polyurethane structures.
To manage that trade-off, the team used a message-passing interface that updated amino acid positions with each pass. The system balanced the optimization of the structure and the binding pocket. The researchers named the resulting software GRASE, short for graph neural network-based recommendation of active and stable enzymes.
This is the important shift in the work. The software was not only comparing protein shapes. It incorporated information related to function, including stability and amino acids likely to interact with the material being digested.
The designed enzyme outperformed natural candidates
The results were strong. Of the 24 most highly rated proteins evaluated by the software, 21 showed some catalytic activity. Eight performed better than the best previously known enzyme. The top design had 30 times the activity of that earlier enzyme.
The performance improved further when the researchers added diethylene glycol and heated the mixture to 50° C. Under those conditions, the designed enzyme was over 450 times as active as the best-performing natural enzyme.
In 12 hours, it broke down 98 percent of the polyurethane in the reaction mixture. The enzyme was also stable enough to be given a fresh mixture of polyurethane two additional times before its activity began to wear out.
The work was not limited to small tests. When the researchers moved to kilogram-scale digestion, 95 percent or more of the material was broken down into the starting materials used to make the polyurethane.
- 15 reported polyurethane-degrading enzymes were tested at the start.
- 24 highly rated designed proteins were evaluated by the software.
- 21 showed some catalytic activity.
- Eight beat the best previously known enzyme.
- The best design reached 30 times that enzyme’s activity before the diethylene glycol and 50° C condition.
- With diethylene glycol at 50° C, it was over 450 times as active as the best-performing natural enzyme.
What this points to
The immediate subject is polyurethane recycling, but the broader implication is about protein design. The researchers emphasize that their tools look beyond a protein’s 3D structure alone. They also consider functional details that affect whether the protein can actually do the job.
That distinction matters for plastic waste because each polymer can pose a different chemical challenge. Finding an enzyme for one material does not automatically solve another. A more targeted design process could help researchers work through these separate problems with more precision.
The source article does not claim that polyurethane waste has been solved. It describes a designed enzyme that works with an industrial-style process and can break the polymer into reusable building blocks under tested conditions. That is a narrower claim, but an important one.
If polyurethane can be digested into useful starting materials rather than converted into a chemical mixture bound for hazardous-waste incineration, the recycling pathway changes in kind. The material is no longer just being degraded. It is being pushed back toward the chemistry needed to make fresh polyurethane.