How AI Helps Stabilize Coronavirus Spike Proteins for Universal Vaccine Design
Article audio
Listen to a generated narration of this post.
Coronaviruses have repeatedly jumped from animals to humans, causing outbreaks that remind us how vulnerable we remain to viral spillovers. Central to these viruses’ ability to infect cells is the spike protein, a complex molecular machine that changes shape to fuse virus and host membranes. But this spike protein is inherently unstable in its prefusion form—the shape that vaccines need to target. Now, scientists have harnessed artificial intelligence to engineer a more stable version of the spike from a common human coronavirus, opening the door to vaccines that could protect across related viruses.
TL;DR
- The coronavirus spike protein’s prefusion form is metastable and difficult to produce for vaccines.
- Using the AI tool ReCaP, researchers designed amino acid substitutions that stabilize the OC43 coronavirus spike, with changes transferable to other embecoviruses.
Coronaviruses are a diverse family of RNA viruses infecting many mammals and birds, including humans. The Embecovirus subgenus includes human coronaviruses OC43 and HKU1, which usually cause mild respiratory illness but can be dangerous for vulnerable people. They also include animal viruses like equine coronavirus (ECoV), which affect horses. The spike protein on the virus surface is key to infection and the main target for vaccines. However, the spike’s prefusion conformation—the shape it holds before fusing with a host cell—is unstable and tends to change shape or degrade, complicating vaccine development. Stabilizing this form is crucial for creating effective vaccines that induce protective immunity.
The researchers started with the OC43 spike protein as a prototype. They combined phylogenetic analysis of OC43 strains, structural insights, and an AI-driven algorithm called ReCaP to predict amino acid changes that would enhance stability. They tested these substitutions individually and in combination in lab-grown cells, measuring protein expression levels and thermal stability. They also designed constructs to improve trimerization—the assembly of three spike units—without relying on foreign protein domains that might provoke unwanted immune responses. To understand how their changes worked, they used cryogenic electron microscopy (cryo-EM) to visualize the stabilized spike structures at high resolution, including the first-ever structure of the equine coronavirus spike.
The team identified specific amino acid substitutions in the spike’s stem region and other domains that significantly increased expression and thermal stability. These changes effectively locked the spike in its prefusion form by preventing the release of the fusion peptide and keeping the receptor-binding domain in a ‘down’ conformation. Remarkably, the substitutions designed for OC43 were transferable to spikes from related embecoviruses, including equine coronavirus and HKU1, demonstrating a broadly applicable stabilization strategy. The cryo-EM structures confirmed that the engineered spikes maintained the desired prefusion shape and revealed how improved polar interactions and fatty acid binding at the interfaces contributed to stability.
This work provides a generalizable framework for stabilizing embecovirus spike proteins, a crucial step toward developing universal vaccines that can protect against multiple related coronaviruses. By leveraging AI to rationally design stabilizing mutations, the researchers overcame a longstanding challenge in coronavirus vaccine development. The ability to transfer these stabilizing features across species highlights the potential to prepare vaccines not only for known human viruses but also for veterinary pathogens and future spillover threats. This approach supports pandemic preparedness by enabling the design of broadly protective coronavirus vaccines.
While the AI-guided design improved spike stability and expression in laboratory settings, further studies are needed to evaluate how these stabilized spikes perform as vaccine antigens in animal models and humans. The immune response elicited by these engineered proteins, including the breadth and durability of protection, remains to be tested. Additionally, although the substitutions were transferable among embecoviruses studied, their applicability to more distantly related coronaviruses requires further investigation. Finally, the long-term effects of using such engineered proteins in vaccines, including potential off-target immune responses, will need careful assessment.