AI assisted design of ligands for Lipocalin-2

Abstract

Lipocalin-2 (LCN2) is an acute-phase glycoprotein whose upregulation is linked to blood- brain-barrier breakdown and neuroinflammation, making it an attractive diagnostic and therapeutic target. We developed an end-to-end, AI-guided workflow to rapidly design de-novo miniproteins that bind LCN2. Backbone scaffolds were generated with RFdiffusion, sequences were optimized with ProteinMPNN, and candidates filtered in silico using a consensus of AlphaFold2 confidence metrics (mean interface pAE < 10) and binding free energy predicted by Prodigy. From an initial library of 10,000 designs, five were expressed and purified from E. coli. Using biolayer interferometry (BLI) we identified MiniP-2 as the lead construct, exhibiting a dissociation constant (Kd) of 4.2 nM. Structural modeling revealed that binding is primarily mediated by backbone hydrogen bonds along with a stabilizing salt bridge between Arg37 of MinP-2 and Asp97 of LCN2. These findings demonstrate that a fully computational generative workflow can yield nanomolar LCN2 binders in a single design-build-test cycle. MinP-2 represents a promising starting point for affinity maturation, structural studies, and in vivo evaluation as an imaging probe or antagonist of LCN2-mediated signaling. Specifically, SPR competition experiments showed that MinP-2 can inhibit LCN2 binding to MMP-9, suggesting its potential to mitigate the pathological effects of this interaction within the central nervous system.