Discovery of thiol-reactive Staphylococcus aureus sortase A inhibitors from natural products by machine learning and thiocholine mass-shift profiling.

Methicillin-resistant Staphylococcus aureus (MRSA) remains a major clinical challenge, and Sortase A (SrtA), a cysteine transpeptidase involved in surface-protein anchoring and biofilm formation, represents an attractive anti-virulence target. In this study, we developed an integrated analytical workflow combining machine-learning (ML)-guided taxon prioritisation, affinity magnetic capture, mass-shift profiling, and biochemical target-engagement assays to identify natural-product SrtA inhibitors from complex botanical extracts. A random-forest classifier trained on ChEMBL SrtA bioactivity data was applied to the LOTUS natural-product database to prioritise Michael-acceptor-containing botanical sources, leading to the selection of Piper retrofractum for constituent-level analysis. Subsequent chemical profiling and affinity-capture experiments identified piperlongumine as a thiol-reactive constituent preferentially associated with active SrtA. Piperlongumine inhibited SrtA transpeptidase activity with an IC50 of 27.83 ± 1.04 μmol/L and showed a higher TCI-adduct response and stronger SrtA inhibition than the structurally related alkaloid piperine. Time-dependent enzyme inhibition, competitive 5-FAM-LPETG bead imaging, and mPEG-maleimide thiol-occupancy analysis further supported thiol-dependent SrtA engagement under the assay conditions. Piperlongumine also inhibited biofilm formation across multiple S. aureus strains while exerting limited effects on planktonic growth at corresponding concentrations. These findings demonstrate an analytical strategy that links computational prioritisation, crude-extract chemical profiling, affinity capture, and target-engagement validation for the discovery of thiol-reactive natural products targeting SrtA.