Probing Severing Enzyme's Functional States with Molecular Simulations and Machine Learning Approaches

Severing enzymes, such as spastin, are microtubule (MT) associated proteins that interact with the MT lattice by removing tubulin and causing internal breaks for regulating MT's various functions. Spastin is an ATPase that forms homohexameric states in the presence of ATP and tubulin carboxy-terminal tails (CTTs), which protrude from the surface of the MT. Elucidating the main states of spastin and their allosteric networks, which are responsible for the function of this machine, along with spastin's interaction with the MT lattice are still outstanding problems, given the size and the complexity of the machine and its substrate. Here, we built a Markov State Model of all-atomistic simulations of the spastin motor in order to identify kinetically relevant protein conformations. Using biochemical descriptors, we then characterized each distinct conformation and applied machine learning classification algorithms to attribute descriptor differences to specific allosteric sites, which can be compared to experimentally determined allosteric networks. Our coarse-grained studies of one of these spastin conformations on a MT lattice yielded tubulin extraction pathways as a function of spastin's orientation relative to the lattice and the binding strength. Taken together, our studies can collectively enhance our understanding of the severing mechanism and its dependencies.