Just about a month ago, Pfizer scored its second huge win of the pandemic when the U.S. Food and Drug Administration issued another emergency use authorization to the company—this time for Paxlovid, a Covid-19 therapeutic shown to produce a reduction in risk of hospitalization or death by up to 88 percent if taken soon after the onset of symptoms. Within days, the federal government had ordered 20 million courses of Paxlovid, and Pfizer is expecting to make 80 million courses by the end of the year.
When Paxlovid was still being vetted, a team of researchers from the University of Valencia applied supercomputing power to elucidate how, exactly, the drug—formally known as nirmatrelvir (or the even less catchy “PF-07321332”)—inhibits SARS-CoV-2. HPCwire spoke to Iñaki Tuñón, a professor of physical chemistry at the University of Valencia, about the research that he and his colleagues conducted on Paxlovid.
The basics of Paxlovid
Paxlovid works by binding to SARS-CoV-2’s 3CL protease (3CLpro), an enzyme that serves as a crucial piece of the virus’ replication process.
“When SARS-CoV-2 infects a human cell, [it] utilizes its transcription machinery to translate the viral genome into two long polyproteins that must be cleaved to produce the non-structural proteins that the virus needs,” Tuñón explained. “This key function is mostly performed by 3CLpro. This enzyme cleaves the polyprotein at 11 positions, targeting a sequence preference that is not used by human proteases. So, this enzyme is an attractive target for inhibitor design, without remarkable side effects in the host.”
Simulating the antiviral
So, when Tuñón’s team set out to understand Paxlovid’s function and efficacy, they set a laser focus on 3CLpro, simulating the enzyme alongside its natural substrate as compared to its interactions with several inhibitors designed to target it, all within the physiological constraints of the human body.
For these simulations, Tuñón said, they used a computational hybrid methodology which combines classical molecular dynamics with quantum mechanics, in particular a density-functional theory method. And to power those simulations, they turned to their colleagues up the coast at the Barcelona Supercomputing Center (BSC), who provided access to their MareNostrum 4 supercomputer.
MareNostrum 4’s main block of 3,456 nodes is powered by Intel Xeon CPUs, and that system delivers 6.5 Linpack petaflops, landing it 74th on the most recent Top500 list. However, the system also includes three so-called emerging technology clusters: one based on IBM Power9 CPUs and Nvidia Volta GPUs; a second based on AMD Rome CPUs and AMD Radeon Instinct MI50 GPUs; and a third based on 64-bit ARMv8 CPUs. (MareNostrum 4’s successor, the pre-exascale MareNostrum5 system, is ostensibly forthcoming and is intended to continue this focus on emerging technologies.)
“All simulations were performed in MareNostrum,” Tuñón said. “Classical molecular dynamic simulations were performed in a machine node that contains one Nvidia Volta GPU and IBM Power9 processor with 40 cores in total. Every simulation of one microsecond took 156 hours to be completed. As three replicas were made, the total computational time for each system was 468 hours.”
“Regarding the computational resources for hybrid quantum/classical simulations, 96 machine nodes were used,” he continued. “4608 threads were used during 432 hours to get the final results.”
What the simulations revealed
Those results illuminated exactly why Paxlovid has proven so effective in debilitating SARS-CoV-2.
“Analyzing the interactions of the inhibitor by residue, we have observed that there are regions in which the inhibitor can almost reproduce the behavior of the natural substrate,” Tuñón said. “During the chemical process, Paxlovid has the advantage of having a small warhead that decreases the conformational freedom shown by other inhibitors reported previously. From the experimental point of view, Paxlovid has shown a high efficiency not only with the 3CLpro enzyme of the original variant of SARS-CoV-2, but also with the mutation associated [with the] Omicron SARS-CoV-2 variant.”
The researchers also produced a visualization of Paxlovid inhibiting the virus, which can be viewed below.
These simulations, of course, aren’t just for fun—understanding how Paxlovid works against 3CLpro at a minute, precise level could have meaningful implications for how Paxlovid and similar therapeutics are created and iterated upon as the pandemic progresses. “Simulations [offer] a quantification of the interactions established between the drug and the enzyme, [providing] valuable information about possible strategies for future improvements,” Tuñón explained.
Indeed, even as the Omicron variant appears to have begun descending from its peak in many areas, Tuñón and his team aren’t slowing down.
“We are working in two different directions,” he said. “On one side, we would like to simulate the behavior of other inhibitors derived from Paxlovid, to see if we are able to propose chemical modifications to improve its efficiency as an antiviral.” On the other side of things, he continued, the team is working to predict mutations that might stunt the efficacy of Paxlovid in future variants.
“Will SARS-CoV-2,” he asked, “be able to escape from antiviral treatments?”