Machine learning reveals signatures of promiscuous microbial amidases for micropollutant biotransformations

Abstract

Organic micropollutants - including pharmaceuticals, personal care products, pesticides and food additives - are prevalent in the environment and have unknown and potentially toxic effects. Humans are a direct source of micropollutants as the majority of pharmaceuticals are primarily excreted through urine. Urine contains its own microbiota with the potential to catalyze micropollutant biotransformations. Amidase signature (AS) enzymes are known for their promiscuous activity in micropollutant biotransformations, but the potential for AS enzymes from the urinary microbiota to transform micropollutants is not known. Moreover, characterization of AS enzymes to identify key chemical and enzymatic features predictive of biotransformation profiles is critical for developing benign-by-design chemicals and micropollutant removal strategies. In this study, we biochemically characterized a new AS enzyme with arylamidase activity from a urine isolate, Lacticaseibacillus rhamnosus, and demonstrated its capability to hydrolyze pharmaceuticals and other micropollutants. To uncover the signatures of AS enzyme-substrate specificity, we then designed a targeted enzyme library consisting of 40 arylamidase homologs from diverse urine isolates and tested it against 17 structurally diverse compounds. We found that 16 out of the 40 enzymes showed activity on at least one substrate and exhibited diverse substrate specificities, with the most promiscuous enzymes active on nine different substrates. Using an interpretable gradient boosting machine learning model, we identified chemical and amino acid features predictive of arylamidase biotransformations. Key chemical features from our substrates included the molecular weight of the amide carbonyl substituent and the number of charges in the molecule. Important amino acid features were found to be located on the protein surface and four predictive residues were located in close proximity of the substrate tunnel entrance. Overall, this work highlights the understudied role of urine-derived microbial arylamidases and contributes to enzyme sequence-structure-substrate-based predictions of micropollutant biotransformations.