Speaker
Description
Per- and polyfluoroalkyl substances (PFAS) belong to a family of more than 4000 compounds known as “forever chemicals”; the high stability of C–F bonds hinders their degradation, hence their extensive spread and presence in the environment. PFAS have been related to a broad spectrum of medical conditions such as hypothyroidism, decreased antibody levels in neonates, or endometrial dysregulation; evincing the complexity of their chemistry and the emerging threat they pose. As an initiative to address this growing concern, we designed dehalogenase variants with REXzyme, an encoder-decoder transformer model that generates enzymes for user-defined chemical reactions. 119 designed sequences were expressed and purified in a high-throughput manner with an Opentrons Flex robot to test their activity against both chlorinated and fluorinated species of AcOH. Five candidates were active against FAcOH, and from those candidates two of them were also active against F2AcOH. In addition, eighteen were active against ClAcOH, and from those candidates nine of them were also active against Cl2AcOH. Among all these candidates, two of them stood out because they were active against all four compounds, compared to the natural dehalogenase Rha0230 from Rhodococcus jostii, which is only active against monohalogenated species. Such candidates were both analysed by SAXS and subjected to screening of crystallization conditions. SAXS proved they behave as globular dimers in solution. Regarding the crystallization screening, only one candidate rendered well-diffracting crystals among several conditions, reaching 1.8 Å at the outer resolution shell. Data processing and integration with XDS was followed by Molecular Replacement (MR) in Phenix using its AlphaFold prediction. Surprisingly, MR failed when the search was done with the predicted dimer due to clashes, whereas two iterative searches with the predicted monomer provided a successful initial model. Refinement of the working model achieved Rfree/Rwork=0.2314/0.1992 with no Ramachandran outliers, yet no fluoride coordination was observed in the catalytic centre. A potential strategy could be soaking the crystals with the substrate and perform time-resolved crystallography, which may capture the different steps of the catalytic reaction. Such an approach would let us discern whether the better activity of our design relies on substrate recognition and reaction mechanism or structural reorganization of the backbone to enhance the catalysis.