March 24 — Influenza is more than a seasonal nuisance leading to a few days of discomfort and a brief absence from school or work. According to the World Health Organization (WHO), the disease is responsible for hundreds of thousands of deaths each year and has the potential to mutate into a more virulent, contagious form that claims millions of lives (as happened in the 1918 “Spanish flu” pandemic).
To better understand the influenza infection process and to explore novel opportunities for drug and vaccine development, the research team led by Rommie Amaro at the University of California, San Diego, constructed an atomic-resolution model of the entire influenza viral coat and simulated this 160-million-atom system on the Blue Waters supercomputer at NCSA.
“Blue Waters really made simulation on this scale possible. It was a critical resource for us,” says Jacob Durrant, a post-doctoral researcher in Amaro’s lab and at the National Biomedical Computation Resource, who is currently analyzing the simulation data.
Amaro’s lab focuses on developing computational methods and applying them to questions of biophysics and drug discovery. In the case of influenza, large-scale computational models are useful because electron microscopy and X-ray crystallography lack either the atomic resolution or size scaling necessary to answer a number of important questions. Furthermore, lab accidents involving “gain-of-function” experiments, which try to anticipate what new live viral strains will evolve in nature by artificially evolving those strains in the lab, pose the frightening risk of releasing artificially virulent and contagious viruses “into the wild,” generating the very pandemic that researchers seek to anticipate and prevent. Computational modelling may reduce the need for these risky experiments.
Proteins Key to the Infection Process
The team’s recent simulations on Blue Waters focused on the influenza viral coat. This outer layer of the virus includes two spike-like glycoproteins (neuraminidase and hemagglutinin) that play vital roles in both the initial and final stages of influenza infection. First, hemagglutinin latches onto molecules on the host cell’s surface, creating a link that enables the virus to enter the cell and reproduce. The replicated viruses then bud from the infected cell but remain attached by the same kinds of molecular links that connected the original invading virus. Neuraminidase cuts those links so the newly spawned viruses can move on to infect other host cells.
The various components of the viral coat have been studied in isolation, which has led to anti-flu drugs such as Tamiflu™. But as the influenza virus mutates and develops resistance to these approved drugs, new therapeutics must be developed. Amaro and Durrant hope that modelling the entire virus coat and observing how its various components interact with one another and with their microscopic environment will yield important new insights.
Creating the Viral Coat Model
The construction of this complex and detailed model was aided by the work of Amaro’s collaborator Alasdair Steven at the National Institutes of Health, who used electron tomography to determine the general shape of the influenza virus and the approximate locations of the glycoprotein spikes. Next Amaro’s team transformed this low-resolution data into a high-resolution atomistic model. Durrant used LipidWrapper, software he developed that can wrap lipid-bilayer models around any surface regardless of its geometry, to wrap an expansive lipid-bilayer model around the entire volume of the virus. He then positioned atomic-resolution models of the glycoproteins at the appropriate locations.
“One of the really exciting aspects of this research is the development of an integrative structural model,” Amaro says. “You have to bring together many different pieces of data to create these realistic models.”
“With these platforms, we can create any strain of the virus that we want.”
Modeling Millions of Atoms on Blue Waters
When immersed in a bath of virtual water with the appropriate electrolytes, the viral-coat system contained 160 million atoms. Drawing upon more than 114,000 CPUs at a time on the Blue Waters supercomputer, Amaro’s team simulated 158 nanoseconds of the coat using NAMD, a molecular dynamics program developed by the Theoretical and Computational Biophysics Group at the University of Illinois at Urbana-Champaign.
“These new capabilities provided by Blue Waters allow us to simulate at relevant spatial and temporal scales,” Amaro says. “It allows us to ask questions and test new hypotheses that no one has been able to explore before.”
Since completing the viral-coat simulations in January 2016, the team has moved into data analysis. Durrant visited the Theoretical and Computational Biophysics team (also designated the Center for Macromolecular Modeling and Bioinformatics by the National Institute of General Medical Sciences) in February, receiving their support in the investigation.
“We’re very interested in how drugs might bind to these different proteins,” Durrant says. “Drugs typically only bind to proteins if they can fit snugly into small pockets on the protein surface, but in real life the shapes of these pockets are constantly changing. Designing a drug is kind of like trying to hit a moving target. Which of the many pocket shapes is most complementary to the given drug you’re studying?”
“One of my theories is that the surrounding environment might impact the way these flexible glycoproteins change their shapes,” Durrant says. “You might not see certain pharmacologically relevant binding-pocket shapes if you only simulate the proteins in isolation.”