Molecular Recipe for γ-Secretase Modulation from Computational Analysis of 60 Active Compounds

gamma-secretase is a membrane protease complex that catalyzes the cleavage of the amyloid precursor protein to produce the infamous A beta peptides involved in Alzheimer's disease (AD). Major efforts aim to modulate this cleavage to reduce the formation of longer, more toxic A beta peptides, yet the molecular basis of this modulation remains unknown. We studied the quantitative structure-activity relations using a carefully curated data set of 60 experimental EC50 values (the GSL60 data set). To ensure adequate optimization, we used 10 different methods to build the models, Y-randomization, 10-fold repeated cross-validation, and explicit external validation on a secondary data set. Neural network optimization best reproduced experimental log EC50. We find that only four descriptors, the number of hydrogen-bond acceptor sites, the topology of the drug, the dehydration energy, and the binding energy to gamma-secretase, define most of the potency of gamma-secretase modulators. We explain this as a compromise between the binding free energy to the protein and required hydrogen bond networks in the actual modulatory sites. Our model suggests that many molecules can modulate cleavage simply by contributing their binding energy to stabilize the compact ternary complex with C99. This result is in line with a mechanism, referred to here as FIST ( Fit, Stay, Trim), where stronger binding to the semiopen state leads to longer retention time and maximal C99 trimming to produce shorter innocent A beta peptides, whereas AD-causing PSEN1 mutations favor the open state by reducing hydrophobic packing, retention time, and trimming and modulators strengthen interactions in the ternary complex to increase the C99 retention time and trimming, ultimately producing more short, nonpathogenic A beta peptides. Our results may aid the development of new gamma-secretase modulators with optimal hydrogen bonds, shape, and hydrophobicity but more importantly provide a structural-chemical model of the modulation of A beta production.