Abstract 2360 Identifying Neuroprotective Small Molecules Targeting Membrane-Bound Tau Oligomers with Physics and Machine Learning-based Models

The intracellular neurofibril entanglements of protein tau are well-known histological markers of Alzheimer's disease. However, recent in vivo and in vitro studies suggest that membrane-bound tau oligomers, which are much smaller and less ordered than tau fibrils, can be more cytotoxic to brain neurons. Aging, tau aggregation, and oxidative stress are key risk factors contributing to the early progression of Alzheimer's. To date, tau fibrils have been the main therapeutic target studied, with recent virtual drug discovery studies focusing on tau fibril disaggregation and elimination of reactive oxygen species in early stages of Alzheimer's. Thus, various tau fibril-targeting molecules have been proposed, like the natural molecule EGCG (epigallocatechin-3-gallate), which is found in green tea and has both tau-disaggregating and antioxidant functions, and the drug-like molecule CNS-11, which resembles EGCG and has tau-disaggregating but no antioxidant functions. Whether or not those tau fibril-targeting molecules are also effective against the more cytotoxic membrane-bound tau oligomers is not known. Using coarse-grained (CG) and atomistic (AA) molecular dynamics simulations, we have successfully designed and simulated various self-assembled tau oligomers of sizes up to tetramers in solution. The soluble oligomers bind to mimics of the cytoplasmic and exoplasmic leaflets of the neuronal plasma membrane containing phosphatidylserine (PS) and ganglioside (GM1) lipids, respectively, within a few microseconds. The simulated membrane-bound tau oligomers were stable for at least 5 microseconds in the CG time scale and exhibited surface-induced protein folding on both PS- and GM1-containing membranes. Using AutoDock Vina, a molecular docking program, and a recent machine-learning (ML) docking-pose-rescoring algorithm, SCORCH (Scoring Consensus for RMSD-based Classification of Hits), we performed virtual docking and rescoring of EGCG and CNS-11 binding to our in silico membrane-bound tau oligomers of various sizes, as well as to experimental tau fibril structures as controls. SCORCH, trained on datasets containing experimental protein-ligand complex structures and docking decoys, provides improved pose identification of ligand-oligomers and ligand-fibril complexes. Our results reveal that both EGCG and CNS-11 bind to the inter-chain regions of the tau oligomers in all membranes studied. Interestingly, the vina binding affinity and SCORCH score of both EGCG and CNS-11 increases with the oligomer size for the exoplasmic GM-containing membrane. However, this trend in the SCORCH score reverses for the cytoplasmic PS-containing membrane. As observed for tau-fibrils, several lysine and histidine residues are predicted to participate in the binding of EGCG and CNS-11 to membrane-bound tau oligomers. Our results address important pathophysiological pathways of tauopathies involving intracellular and extracellular regions of neuronal plasma membranes linked to Alzheimer's. The combination of physics- and machine-learning-based computational tools provides new insights into membrane-bound tau-oligomers, which could prove useful for the design of novel therapeutics. Acknowledgment - This work was supported by the Welch Foundation (W-2057-20210327), NSF (OAC 1531594), and NIH (5SC3GM140980-03)