AI-guided identification of a covalent SMAD3 regulatory mechanism of celastrol in diabetic kidney disease

Abstract

Renal fibrosis is a defining pathological feature of diabetic kidney disease (DKD) and a major contributor to progressive renal dysfunction. Although aberrant activation of transforming growth factor-β (TGF-β)/SMAD3 signaling is a central driver of renal fibrogenesis, effective targeted therapies remain limited. Here, using an AI-guided target-prioritization strategy based on the deep learning platform Uni-Mol, we selected Celastrol for experimental investigation and subsequently characterized its molecular interaction with SMAD3. Activity-based protein profiling combined with a competitive iodoacetamide-based probe assay identified SMAD3 as a direct covalent target of Celastrol, which was further supported by molecular docking and site-directed mutagenesis. Mechanistically, Celastrol preferentially engages the non-phosphorylated form of SMAD3 and promotes KPNB1-dependent nuclear translocation, resulting in the premature nuclear accumulation of functionally inactive SMAD3 and reduced association of SMAD3 with TGFBR1. In parallel, Celastrol disrupts the interaction between SMAD3 and SMAD4, thereby limiting the formation of transcriptionally active SMAD complexes and attenuating profibrotic signaling. Collectively, our findings identify SMAD3 Cys421 as a functionally relevant covalent site for Celastrol engagement and reveal a previously uncharacterized mechanism by which Celastrol regulates the spatial and functional state of SMAD3. These findings provide mechanistic insight into the antifibrotic activity of Celastrol and may inform the future development and optimization of Celastrol-based therapeutic strategies for renal fibrosis in DKD.