Genomic Science Program
U.S. Department of Energy | Office of Science | Biological and Environmental Research Program

Optimizing Enzymes for Plastic Upcycling Using Machine Learning Design and High Throughput Experiments


Brenna Norton-Baker1 (, Benjamin Fram2,3, Elizabeth Bell1, Richard Brizendine1, Karla Ilic Durdic4, Perry Ellis4, Japheth Gado1, Erika Erickson1, Samuel Lim3, Debora Marks3, John McGeehan1,5, Natasha Murphy1, Nathan Rollins3, Chris Sander3, Nicole Thadani3, David Weitz4, Allison Werner1, Wentao Xu4, Xinge Zhang4, Gregg Beckham1, Nicholas P. Gauthier2,3* (


1National Renewable Energy Laboratory; 2Dana-Farber Cancer Institute; 3Harvard Medical School; 4Harvard University; 5University of Portsmouth


Researchers aim to create new and optimized polyethylene terephthalate-depolymerizing enzymes (PETases) useful for industrial application. [Aim 1] Design novel PETases that are significantly different (25 to 65+ mutations) from known PET-depolymerizing enzymes and contain unique properties useful for performant enzymatic PET recycling and upcycling. Introducing many simultaneous mutations, while maintaining function, will enable researchers to more efficiently search for altered properties that depend on primary amino acid sequence. [Aim 2] Optimize previously described PETases by testing millions of mutagenized variants using directed evolution. Starting with existing functional PETases and exploring small changes in many distinct sequences using a novel ultra-HTP functional assay, researchers will optimize enzymes with improved properties by varying experimental conditions. [Aim 3] Characterize performance metrics of new and optimized PETases in detail including solvent tolerance, stability, catalytic rate, and substrate promiscuity.


Plastic use is ubiquitous in the modern world, and polyethylene terephthalate (PET) is one of the most abundantly produced plastics (and the most highly produced polyester), with ~65 million metric tons manufactured annually. To the consumer, PET is likely most recognizable as the plastic used to make beverage bottles. Like many plastics, traditional mechanical or chemical means of PET deconstruction and upcycling are costly and inefficient. Recently, biological enzymes capable of breaking down PET into its basic building blocks (terephthalic acid and ethylene glycol) have garnered significant attention as an attractive means of dealing with the plastic problem. These enzymes are currently undergoing pilot studies for implementation in enzyme-based recycling. However, there are significant limitations to current enzymes, including the need to perform costly pre-processing of the plastic waste before the enzymes are able to work. Further optimization of these enzymes is necessary to make the process profitable and thereby incentivize commercialization of this biology-based green recycling technology.

In this work, researchers apply recent advances in artificial intelligence and machine learning to design new versions of enzymes capable of breaking down PET. From a multiple sequence alignment of natural homologs of PETase, researchers derive models of the protein family. These models are used to design new sequences with a large number of mutations (5 to 20% of positions) that are in parsimony with the homologs yet are distinct in their primary sequence. In addition to this new computational methodology for sequence design, the team also rationally introduced point mutations known to promote PETase enzyme stability or activity. To test design variants, the team developed a high-throughput robotic testing platform capable of enzyme purification and characterization of hundreds of putative PETase enzymes across a range of 16 conditions (pH, temperature, substrate).

Using the new experimental platform, researchers have now tested ~800 designed PETases that contain ~10 to 70 amino acid changes relative to their wild type chassis. The majority of these enzymes have bona fide PETase activity, deconstructing PET into TPA and MHET. Additional rounds of design on top of these already performant designs tend to further enhance activity. In addition to demonstrating drastically increased activity compared to the wild type from which they were derived, many of the designs also demonstrate higher activity than the top PETases in the literature (e.g., LCC-ICCG). Individual designs tended to be performant in very specific conditions (i.e., amorphous film / 70°C / pH 4.5) rather than showing broad performance across multiple conditions. From this large diverse set of designed PETases, researchers generally have one or more variants that best the top published PETases in every condition. Ongoing work is aimed at probing the mechanistic consequences of both individual and sets of mutations, furthering a predictive understanding of these performant enzymes. The simultaneous application of combinations of designs in future work will probe for synergy and additivity towards breakdown of diverse PET substrates.

Funding Information

This research was supported by the DOE Office of Science, BER Program, grant no. DE-SC0022024.