When SARS-CoV-2 attacks a cell, it does so with its spike proteins – the protrusive, invasive proteins that adorn it, giving the coronavirus its “coronal” (crown-like) appearance. The spike proteins undergo dramatic conformational changes when the virus is binding to a cell, presenting a challenging moving target to the researchers who are attempting to disable the spike protein (and thus cripple the coronavirus). Now, a model running on supercomputers at the Texas Advanced Computing Center (TACC) is aiming to pinpoint the dynamics of the spike protein in order to illuminate those processes for researchers.
The spike protein comprises three parts that move – and even separate – as it opens up the human ACE2 receptor in order to fuse its helical core to and infect the host cell. The research, run by Numan Oezguen (an instructor at the Microbiome Center of Texas Children’s Hospital and at the Baylor College of Medicine), is simulating the spike protein to answer extremely specific questions about its operation.
“Can I see when and how the receptor binding domain of the virus that binds to the ACE2 changes its conformation?” Oezguen asked in an interview with TACC’s Jorge Salazar. “How does it move from its closed, or down, conformation, into the up conformation that is capable of binding to the receptor? And can we disable it?”
To answer these questions through simulation, Oezguen employed a simulation of half a million atoms to fill in the gaps left by the cryogenic electron microscopy models of the coronavirus provided by physical labs. Oezguen ran his model on the Longhorn subsystem of TACC’s Frontera supercomputer.
Frontera’s primary compute system is capable of 23.5 Linpack petaflops, placing it 8th on the most recent Top500 list of the world’s most powerful publicly ranked computers; Longhorn, meanwhile, is capable of around 2.3 Linpack petaflops, placing it 152nd. Longhorn is an IBM Power9 system with 104 nodes, each powered by four Nvidia V100 GPUs.
Oezguen put this GPU firepower to good use, completing over 47,000 node hours of simulations on Longhorn and leveraging its graphics abilities to run AMBER18, a molecular dynamics program. “I’m very happy with Longhorn,” Oezguen said. “Even the very large system with half a million atoms progresses at about 20 million femtoseconds per day. I’m calculating every femtosecond the position and velocity of each atom. This is mind-boggling, actually, how much computation is going on.”
However, the scale of the molecular simulation is still daunting.
“Even with these fast resources,” he continued, “the goal for one microsecond has taken about 50 days. Imagine if you would have less efficient machines to work with. It will take so long you would not be able to finish it in a human lifetime. We’ve come very far, and I’m very grateful to have access to this great resource,” Oezguen said.
The simulations aren’t complete, and Oezguen is eagerly awaiting the appearance of a telltale piece of movement: the priming of the spike protein for infection. If Oezguen is unable to observe this movement, he is worried that environmental factors may be slowing – or even preventing – the process from working.
“Hopefully, I will see the movement and then analyze what regions of the protein are moving first or enabling the movement,” Oezguen said. “Once we know this, then we can think about ways of preventing this. If we can prevent the up movement of the receptor binding domain on the spike, then everything else stops. The virus cannot enter the cell. This is crucial.”
Header image: The spike protein before separation (left) and after separation. Image courtesy of Numan Oezguen.
To read TACC’s reporting on this research, click here.