HIV Protein Switch May Help Virus Squeeze into Host Cell Nucleus

The fight to eliminate AIDS continues. An important step in infection by HIV is insertion of the viral capsid—the inner protein coat containing its genetic material—through the host cell’s nuclear pore. How the virus does this is a puzzle of squeezing a large structure though a small entrance. It’s also a potential target for better therapies. 

Supercomputer simulations revealed how changes in the shape of the HIV-1 capsid protein might help the capsid be more flexible. University of Pittsburgh scientists used Bridges-2 of the Pittsburgh Supercomputing Center (PSC) and Lonestar6 of the Texas Advanced Computing Center (TACC) to help develop their method, which could also be useful for studying flexibility in other important proteins.

Why HIV-1 Remains a Threat

Year 2024 was a mixed bag in the fight against AIDS. Modern antiviral therapies have turned HIV into a chronic, but survivable, disease. Death rates are at a 20-year low. On the other hand, doctors are unlikely to meet their goal of eliminating AIDS as a health threat worldwide by 2030. In some populations, HIV infection is increasing. And scientists still haven't developed a vaccine. That’s why they continue to study AIDS in search for weak spots in the virus’s infection cycle.

HIV inserts its genes into the host cell’s nucleus in an unusual way. It doesn’t import its RNA genetic material piecemeal. Instead, it stuffs its whole wedge-shaped capsid—basically the whole virus minus its outer membrane—through the nuclear pore in one piece.

The HIV-1 capsid protein makes up most of the protein mesh that forms the capsid. It does this by making connections between separate capsid proteins at different scales. First, either six or five copies of the protein link together via their N-terminal domains to form six-sided hexamers or five-sided pentamers. Then, the opposite end of the protein, the C-terminal domain (CTD), links with CTDs of neighboring hexamers or pentamers to connect them and form a mesh that surrounds the genetic material. The wedge of the capsid is kind of like a soccer ball, which also needs hexagons and pentagons to make a curved shape.

The CTD connections between two proteins—called dimers—in neighboring hexamers or pentamers can adopt different shapes that change the angles between those shapes. Before assembling into the capsid, about 85 percent of the CTDs connect in the D1 shape; the rest, the D2 shape.

Scientists suspected that the pointy end of the capsid’s wedge might help it squeeze through the nuclear pore. What they didn’t know was whether the ability of the CTD to shift angles between the hexamers and pentamers might play an additional role in making the capsid more flexible, and better able to deform to push through. One problem was that the D1 to D2 conversion is so fast, and the D2 shape is so outnumbered by D1. Because of this, D2 doesn’t show up well in imaging and is hard to simulate. D2 was basically invisible to both methods.

How Supercomputers Helped

Postdoctoral researcher Darian T. Yang at the University of Copenhagen wanted to know what D2 looks like, and whether the capsid protein’s ability to change shape might give the capsid extra flexibility to squeeze through. He worked as a graduate student during this study with both University of Pittsburgh Professor of Chemistry Lillian T. Chong and University of Pittsburgh Medical Center Rosalind Franklin Chair of Structural Biology Angela M. Gronenborn. 

Yang carried out an exhaustive series of simulations of the capsid protein with the National Science Foundation-funded Bridges-2 of PSC and with TACC's Lonestar6. He got computing time on Bridges-2 via allocations from ACCESS, the NSF’s high performance computing network. The NSF also awarded Yang's team with a Large Resource Allocation on the Frontera supercomputer, of which Lonestar6 is a subsystem.

The simulations would require powerful graphics processing units (GPUs), and plenty of them. Yang’s simulations would require many repetitions to work out the different ways in which the proteins can behave. Bridges-2, with 34 GPU nodes containing a total of 280 late-model GPUs, fit the bill nicely. For comparison, a high-end graphic design laptop typically has two GPUs.

The team compared its simulation results with laboratory experiments using an imaging technology called nuclear magnetic resonance, or NMR, which can track the shape of the protein via a fluorine atom that scientists attached to the natural protein. By going back and forth between real-life results and the behavior of the virtual proteins in the computer, they could be more confident the simulations were capturing reality.

Looking to Lonestar6 for Answers

Yang's team leaned on Lonestar6's graphics processing unit nodes to run and optimize their weighted ensemble simulations of the HIV-1 CA-CTD dimer using the WESTPA software package, which scaled well with multiple GPUs, and they were able to run these simulations across multiple nodes. They also ran enough standard molecular dynamics simulations on these nodes to build the Markov State Model used for simulation analysis and comparison.

"The simulations allowed by these resources were valuable," Yang said. "We were able to test many different weighted ensemble simulation parameters, as well as advanced multi-region binning schemes and on-the-fly trajectory reweighting schemes. My experience with the Lonestar6 system was good, as it was useful to also have the gpu-a100-dev queue for testing and debugging before sending off the larger scale jobs running across multiple gpu-a100 nodes."

Supercomputers for Disease Investigation

The team's simulations, which were the first of their kind, produced switching rates and populations of the D1 and D2 shapes that matched the behavior of the capsid in the lab experiments. The team announced their findings in the Proceedings of the National Academy of Sciences U.S.A. in February 2025.

The results are encouraging because they suggest that simulations can pair with experiments to find new targets for HIV drugs or vaccines. The method should also be useful for scientists studying other biologically and medically important systems.

"Supercomputer simulations offer a powerful way to reveal the atomistic details of critical biological processes, such as those driving HIV-1,” Yang said. “However, their accuracy depends on being firmly grounded in experimental data. By integrating simulations with lab experiments, we can bridge the gap between the high-resolution insights from computation and the often limited but essential real-world observations from the lab. This integrated approach helps refine our understanding of hidden disease mechanisms and brings us closer towards a fully detailed picture of biological interactions."

Adapted from a PSC press release by Ken Chiacchia.