Design of Novel Dehalogenases using Protein Large Language Models

Abstract

Per- and polyfluoroalkyl substances (PFAS) are toxic compounds linked to cancers, infertility, and vaccine resistance at concentrations above 1 part per trillion. Owing to strong carbon-fluorine bonds, they are persistent in the environment, taking centuries to millennia to degrade. As part of the 2024 iGEM competition, several high school students local to the Pasadena area leveraged recently-developed bioinformatic tools and large language models to discover and design novel reductive dehalogenases predicted to degrade perfluorooctanoic acid (PFOA), a long chain PFAS whose manufacture is prohibited, yet still persists in the environment and drinking water. The team identified 68 enzymes with structural similarity to Acidimicrobium sp. Strain A6 RdhA, the only specific known PFAS degrading enzyme in nature, expressed and refolded 5 of these enzymes in addition to 1 rational, and 3 large language model designs. These designs all have diverse sequences yet all are predicted to retain key substrate and cofactor binding pockets. These enzymes will be assayed on the ability to defluorinate PFOA.