Microtubule mysteries revealed

Each day, a human adult loses on average 50 to 70 billion cells, which die from natural causes alone. New cells replace lost ones by the complex process of cell division, which relies on what scientists call molecular machines to transport chemical cargo to where it is needed for reactions that keep us alive. 

Long polymers called microtubules (MT) play a key role in cell division and chemical transport, paving the way for molecular machines and pushing the nucleus apart to split it. Microtubules are dynamic structural polymers composed of protein building blocks called tubulin that add and lose fragments at their ends. 

Supercomputer simulations by University of Chicago and University of Utah scientists have revealed new behavior happening at microtubule tips, the location where they grow or shrink. This basic research in microtubule stability could help lead to better understanding of neurodegenerative diseases such as Alzheimer's and Parkinson’s as well as aid in design of cancer drugs.

“One of the huge mysteries of the microtubule is what goes on at their tips,” said Gregory Voth, the Haig P. Papazian Distinguished Service Professor of Chemistry at the University of Chicago. 

Voth co-authored research published in the Biophysical Journal (January 2025) that investigated the chemical conversion through hydrolysis of the nucleoside Guanosine-5'-triphosphate (GTP) into Guanosine diphosphate (GPD) happening at microtubule tips. This conversion of GTP to GDP at microtubule tips speeds up both their polymerization and depolymerization, collectively known as MT “dynamic instability,” which facilitates its growth or breakdown. 

“It used to be thought that the tips would splay out, like a ram’s horn, only after the microtubule chemically changed GTP into GDP,” Voth explained. “But that's not true now. This really changes the picture. The tips are always more or less splayed out. The challenging question is whether there was a difference in splaying with GTP or GDP at the MT tip, and the supercomputer simulations were able to help reveal that.”

What’s more, Voth and colleagues were able to achieve a two-fold computational speedup in their simulations. This feat generated all-atom molecular dynamics data and fed that to a machine learning method that could carry the simulation all the way through to the relaxed state of the MT tip, 5.875 microseconds of evolution on a system of 21 to 38 million atoms. Their results revealed key but subtle differences in the structures of MT tips depending on whether they are bound with GTP or GDP. 

TACC awarded Voth a Leadership Research Allocation on the National Science Foundation-funded Frontera, the top U.S. academic supercomputer in the country. Voth reported using 56 million Frontera central processing unit (CPU) core hours to generate four microseconds of all-atom molecular dynamics (AA MD) simulations, an unprecedented length of trajectory for such a large system.

Additionally, computational resources were allocated by the University of Chicago Research Computing Center and the National Institutes of Health-funded Beagle3 HPC cluster.

The team’s ‘equation-free’ multi-scale simulation method incorporated machine learning to extend the simulation to 5.875 microseconds, saving 15 million CPU hours over AA MD alone and accessing dynamical behavior that cannot at present be readily seen in the AA MD alone.

“Without Frontera, we could not have done it,” Voth said. “The big data from Frontera feeds the machine learning algorithm that extends its range beyond what the AA MD can do. They work synergistically together to achieve scales.”

Voth contrasts these MT tips methods with prior work completed on Frontera on HIV-1 capsid modeling that employed coarse-grain model methods to show how the deadly virus enters the cell nucleus. Coarse-grain models simplify the resolution scale, and if done well they retain all the key physics of intractable computer simulations.

Like the coarse-grain work done with HIV-1, Voth’s team used state-of-the-art images of microtubules from cryo-electron tomography as snapshots that elucidated critical structural differences and conformational changes between GDP- and GTP-bound microtubules.

“Supercomputers provide the necessary and accurate data for utilizing machine learning to discover new behavior of critically important collections of proteins in your cells,” Voth said. “It's a beautiful story. Starting from the fundamental ‘big iron’ Frontera simulation and then going up in scale. And in the end, we discovered new behavior pertinent to living cells.”

The study, “Data-driven equation-free dynamics applied to many-protein complexes: The microtubule tip relaxation,” was published January 2025 in the Biophysical Journal. The study authors are Jiangbo Wu, Siva Dasetty, Daniel Beckett, Yihang Wang, Weizhi Xue, Andrew L. Ferguson, and Gregory A. Voth of the University of Chicago; and Tomasz Skora, Tamara C. Bidone of the University of Utah. Funding from the Department of Energy supported the research.