At the end of 2019, strange pneumonia cases started cropping up in Wuhan, China. As Wuhan (then China, then the world) scrambled to contain what would, of course, go on to become the worst pandemic in a century, speculation on its origins reached a quick – if hasty – conclusion. The virus, many reasoned, must have come from Wuhan’s wet markets – open-air markets featuring seafood and meat, with Wuhan’s in particular featuring many illicit animals (such as bats and pangolins) known to be major disease vectors, with Chinese bats in particular thought to be the origin of the original SARS virus (now known as SARS-CoV-1).

But what if that’s not what happened? Over the last year, a team of Australian researchers worked to decipher the origins of SARS-CoV-2 using molecular dynamics and cloud-based HPC resources – and the results may lend further plausibility to a less natural scenario for those first COVID-19 infections. In a talk for Virtual ICM Seminars (part of Supercomputing Frontiers Europe 2021), David Winkler (a professor at La Trobe University, Monash University and the University of Nottingham) discussed his team’s work.

Mysterious origins

Over the course of the pandemic – now past the year-and-a-half mark – the so-called “lab leak hypothesis,” arguing that SARS-CoV-2 escaped from the Wuhan Institute of Virology, has gained traction: and, increasingly, legitimate traction. What was once considered a far-flung conspiracy theory (and one that was quickly shot down by the World Health Organization) is being revisited by researchers around the world. Most recently, those flames were stoked by revelations that a group of workers from the Wuhan Institute of Virology fell ill in early November – news that was quickly followed by President Biden ordering the U.S. intelligence apparatus to determine the origins of the deadly virus.

Some, however, have been skeptical of the virus’ origins for much longer. “There are a number of questions that arise,” Winkler said. “How did the disease arise? How did that jump from some other species into men? How did it adapt to men once it was in men?”

At first, he, like many, applied Occam’s razor, suspecting bats. “Bats were probably the most likely candidate, because there’s a bat coronavirus that’s quite similar to the current SARS-CoV-2,” he said. “But there needs to be an intermediate animal to bridge between bats and humans.” Without a viable intermediate animal to serve as a bridge, natural zoonotic transmission would appear far less likely – and lab escape, far more.

Beginning early in 2020, Winkler and others were wrangled by Nikolai Petrovsky, a professor in the College of Medicine and Public Health at Flinders University and chairman and research director of the Australian biotech company Vaxine (which itself has developed an as-yet-unreleased COVID-19 vaccine). “Nick is quite an entrepreneurial guy,” Winkler said, “and he got a team of people together working on molecular dynamics and homology modeling and so forth to try to understand what was going on here.” 

“We wanted to look at this question: where did the virus come from?” Winkler said. “No one really knows. There’s been, I guess, a bit of an active effort to not canvas the escape-from-a-lab situation, but I think we need to look at all possibilities because all of them are feasible.”

The researchers decided to approach the question from the angle of binding affinity – that is, testing how well the spike protein from SARS-CoV-2 binds to the ACE2 receptors of host organisms. This, they reasoned, would indicate how likely it was that the virus infected, say, a bat, and in turn how likely it would be for the infected bat to infect a human.

“What’s the susceptibility of these species that have been implicated as being potential intermediates in the transmission?” Winkler said. “And do we have to worry about other animals like companion animals – dogs, cats, birds – and farm animals – like horses, sheep, chickens – being a reservoir for the disease? And potentially for us to pass the disease to them, and then from them to pass back to us, which would be rather disastrous?”

With these questions in mind, the team got to work.

Working from the ground up

“So, what we wanted to do is to look at the very first interaction of the virus with humans,” Winkler said. There was one issue: at the time, no 3D structure of the spike protein was available. So using a viral structure retrieved from NCBI GenBank Database in January 2020, the team built their own. (Later, when such structures became widely available, theirs held up well.)

The next part of the puzzle was the ACE2 receptor – or rather, ACE2 receptors. “[The] human [ACE2 receptor model] was available, of course, but we wanted to look at a range of other animals,” Winkler said. “So we had to build models of the ACE2 proteins for other species, because there were no crystal structures available.” 

The species at hand included not only humans, bats and pangolins, but also monkeys (due to their similarity to humans), pets (dogs and cats), farm animals (horses and cattle), common lab test subjects (civets, ferrets and mice), snakes (the king cobra) and tigers (some of which had been diagnosed with COVID-19). The researchers built models of the ACE2 proteins for all of them. (Subsequently, some of the crystal structures appeared in the protein data bank and the researchers were able to use those more accurate versions – but they did find their own structures to be highly accurate.)

The HPC of it all

With the structures in hand, the researchers turned to the heavy-duty computational elements of the research: simulating the virus’ spike proteins binding to the slew of ACE2 receptors in a realistic manner that accounted for uncertainties. 

They began with HDOCK, a state-of-the-art protein-protein docking package that performed the initial docking calculations. These were then optimized using the 2020 edition of GROMACS, a molecular dynamics package dating back to the early ‘90s that remains one of most popular tools for MD simulations and has been widely applied during the pandemic to study spike protein interactions.

GROMACS is a heavy-duty tool, and performing the intensive simulations required to test the hypothesis necessitated correspondingly high-powered computing. In that regard, the team was supported by Oracle Cloud, which has a wide range of programs in place to support researchers with computational time and expertise. “We had computational resources very kindly given to us by the Oracle Cloud system and the Oracle corporation,” Winkler said. “They were very generous with their computational time.” Oracle supplied the researchers with (presumably Nvidia-powered) GPU nodes, allowing the team to use the GPU-accelerated version of GROMACS.

First, they ran a production run of 500 nanoseconds, simulating the binding between the spike protein and the human ACE2 protein. Confirming that 50 nanoseconds was enough to capture the convergence process, they then simulated 100 nanoseconds for the remainder of the bindings. These were all put through multiple production runs using various random starting seeds to capture the uncertainties in the calculations and models. (“We did this as carefully as we possibly could,” Winkler stressed. “We ran the calculations multiple times and so forth, trying to estimate the accuracy.”) After all these runs, a GROMACS tool was used to calculate the binding energies.

And the vector is…

“Surprisingly, we found that humans had the highest affinity,” Winkler said. “You would expect the virus – like the flu virus or a coronavirus – to adapt to its host over time and become more tightly bound, but this was a structure that we had from the very, very early part of the pandemic before the virus would have had time to adapt to a human host.”

“Pangolins were a little bit lower,” he continued. “Bat was quite a long way down. That was where people considered the virus originated, and it needed to pass through an intermediate animal, which at this stage most likely to be a pangolin based on these calculations. But there are other things, like [the] snake, which was also considered a potential source of the coronavirus, and it was way, way down on the binding energy – so it seems very unlikely.”

The researchers did their best to vet those affinities, correlating them with infectivity by species where such data was available to establish a “reasonable sort of qualitative correlation between the observed permissivity of infection and degree of infection and the binding energies[.]” And, Winkler said, the available data supported the results.

But those results, he said, lack the kind of decisive answer that might have precluded the lab leak hypothesis.

“We didn’t really expect the human to come out on top,” he said. “Because we would have thought that if the virus had come from a bat, probably that would come out to be the top – possibly a pangolin. So based on this information, you can’t really exclude the possibility that the virus could have escaped from a lab. I’m not saying it did, but those calculations suggest that it could have come from a lab.”

Winkler cautioned that the results – which have remained in preprint for many months after submission to several journals – are not evidence of deliberate manipulation of the virus (so-called “gain of function” research) by Chinese scientists, a farther-flung hypothesis which he said seemed to not have much – if any – supporting evidence. Much more likely, he said, was an accidental release, though the pangolin remained a plausible bridge between bats and humans.

In any case, he said, “further deliberations” were needed to move closer to the truth.

Update: the research has now been published in Scientific Reports. To read it, click here.