Novel Plastizymes: discovery and improvement of plastic-degrading enzymes by integrated cycles of computational and experimental approaches

Modern life generates enormous amounts of plastic waste: 359 million tons of plastics are produced annually worldwide, of which 90% is produced from fossil fuels and 79% accumulates in landfill or in the natural environment. Collectively all these plastics create an environmental hazard. Furthermore, we are losing valuable materials that could be recycled. As Nature did not encounter plastics for most of its evolutionary history, plastic-degrading enzymes with a metabolic role did not exist. However, recent research into communities of bacteria from oceans and wastewater has shown that over the last 50 years some bacteria have evolved enzymes that can exploit this new nutrient. These plastic degrading enzymes, some of which are known as PETases (as they degrade polyesters, PETs), are not very efficient but represent an exciting starting point to discover and engineer more effective enzymes. Furthermore, international 'metagenome' efforts have been capturing vast amounts of bacterial genomic data from these natural environments, which are now available as resources like MGnify at the European Bioinformatics Institute (EBI).
In this project we will use bioinformatics to harvest enzymes from these massive metagenomic databases, by classifying them into functional and structural classes with useful 'promiscuous' chemical activities. We will use state-of-the-art artificial intelligence (AI) and machine learning (ML) tools to do this, proven to classify families of proteins with high functional similarity. Putative novel plastic-degrading enzymes identified by this approach will be further analysed by ML tools which screen for predicted solubility. We will also perform chemical studies to assess improvements in enzyme activity, compared to the existing, inefficient, PETases. Any putative plastic-degrading enzymes will then provide a starting point for directed evolution experiments where we select new variants of the enzymes with improved properties. To best explore evolution of plastic degrading ability we will use our unique ultrahigh-throughput assay for particle breakdown, with a throughput of over 10 million clones per day. We can thus directly assess the ability of enzymes to chemically act on plastic particles (rather than substrates that only mimic plastics).
This will revolutionise the field of enzymatic plastic degradation, because so far only marginal improvements have been possible using proxy substrates. In addition to efficient screening, the analysis of the output sequences of screening will be fed back into our bioinformatic analyses and target selection. We will also structurally characterise these enzymes to discover how changes in their functional sites have improved their ability to bind and digest plastics. This data will provide detailed insights on how protein sites can diverge and evolve better plastic degrading properties, thus improving our in silico selection protocol. We have performed pilot work on PETases and will build on this and extend to other plastic degrading enzymes (plastizymes). This close integration of 'dry' data science and 'wet' experimental work results in powerful cycles of in silico analysis, experimental tests and refinement of analysis tools that are more powerful than current small scale protein engineering campaigns. The project thus addresses one of the most important (and also most difficult) environmental challenges, but more generally, also provides a paradigm to demonstrate how an interdisciplinary approach can accelerate evolution in cases where no effective natural enzyme is available. If successful, this paradigm would form the basis not just for the 'rules of life' (as mentioned in the call text), but for 'rules beyond life' (as it exists now), targeted to address the future needs of our society.

Our multidisciplinary approach will identify novel plastizymes from MGnify and experimentally test for degradation of various plastics (PET, PA, PCL, PU). We will use metagenomic and metatranscriptomic assembly to provide a comprehensive set of target enzymes, retrieved from a broad range of environments. We will investigate k-mer based methods and develop new signature-based searches of unassembled datasets, to enable data set selection and targeted, gene-specific assembly. We will develop more scalable web-based searches and data access and explore the utility of combinations of enzymes for plastic degradation. We will use publicly endorsed methods for functional family classification to detect plastizymes in MGnify and develop powerful AI and machine learning (ML) strategies that use sequence and structure embeddings to distinguish clusters of plastizymes likely to have distinct binding and activity properties. A novel predictor for protein solubility will also be built. We will carry out dir. evolution experiments (from epPCR and InDel libraries) to improve plastic-degrading enzymes at ultrahigh-throughput (>10e6/day) in microfluidic droplets, including detection of enzymatic breakdown of actual plastic particles by light scattering. Candidates include promiscuous esterases obtained in functional metagenomics droplet screens. Evolution will be mapped by UMIC-seq based on thousands of full-length nanopore sequences over multiple rounds as input data for AI/ML analysis, to go beyond experimentally screened mutants. Expression constructs will be designed for target enzymes with AF2 models as guides and with codon optimisation for E. coli. A panel of solubility-fusions will be screened in E. coli and in V. natrigiens. Analysis of oligomeric state and stability of purified proteins will lead to structure determination by molecular replacement using AF2 models. Soaking crystals with substrate analogues or by probing with fragment libraries will map active sites.