AI-Driven Discovery of a Hydrophobic-Tag Degrader Targeting NSD3 for Lung Squamous Carcinoma Therapy

Nuclear receptor-binding SET domain protein 3 (NSD3) is a histone H3K36 methyltransferase implicated in lung squamous cell carcinoma, yet chemical targeting is challenging due to the shallow PWWP reader pocket. Here, we report XSY12, a hydrophobic-tag degrader of the NSD3 PWWP domain that enables cellular NSD3 depletion (DC50 = 2.21 μM; Dmax = 80.2%). AI-guided discovery using a 3D fingerprinting platform (TF3P) identified a new NSD3-PWWP chemotype, which was optimized via deep learning-based molecular generation to the high-affinity ligand SYC2 (KD = 0.07 μM) and subsequently converted into XSY12. XSY12 promotes proteasome-dependent NSD3 depletion accompanied by HSP90-associated signatures, reduces H3K36 methylation, and induces apoptosis and G2/M arrest in NSD3-dependent models. In vivo, XSY12 achieved measurable exposure and significant tumor growth inhibition in an LUSC xenograft model at 100 mg/kg with acceptable tolerability. Collectively, these results provide a practical workflow linking AI-guided ligand discovery to functional degrader development.