Anthropic maps first 3D structure of a bacterial vault particle

A decades-old biological mystery

Vault particles have puzzled biologists for nearly four decades. These hollow, barrel-shaped objects are among the largest protein complexes found in human cells, yet their specific biological purpose remains a mystery. While they are widespread in complex organisms, researchers have yet to determine what they actually do.

Scientists have proposed several theories regarding their function. Some researchers suggest they might be involved in cellular transport, drug resistance, or immune defense. However, experiments have so far failed to establish any of these roles conclusively.

The structures appear across a wide range of the evolutionary tree, including organisms like slime molds and humans. Because common laboratory organisms often lack vaults, it has been difficult for researchers to study them effectively. This discovery provides a new path forward by identifying related structures in bacteria and bacteriophages.

Mapping the bacterial vault

Anthropic’s biology research team set out to find vault-like proteins in non-mammalian life. They searched approximately 15 billion predicted protein sequences to find distant relatives of known vault proteins. This search identified potential candidates across 29 major bacterial groups and four groups of archaea.

The team selected 44 candidates for laboratory testing and successfully produced 15 proteins. Using cryogenic electron microscopy, or cryo-EM, they confirmed the structures of four candidates. Two of these originated from bacteria, while the other two came from bacteriophages, which are viruses that infect bacteria.

The researchers focused on a phage-derived protein called L05 for detailed analysis. This protein assembled into a complete structure containing 96 copies of itself. The resulting particle measures about 54 nanometers long and 45 nanometers wide, with a molecular mass of roughly 9.5 megadaltons.

Volume and evolutionary clues

The bacterial vault structure revealed some surprising physical characteristics. Despite being shorter than mammalian vaults, the L05 structure encloses approximately 25 percent more volume. This suggests that different types of vaults can achieve different internal capacities through different arrangements.

To understand how these structures grow, the researchers modified the L05 protein by adding two repeating segments. This modification produced an even larger particle, increasing the enclosed volume by approximately 20 percent. These experiments suggest that protein duplication could have helped vaults evolve into various sizes over time.

The team also observed how small structural changes can prevent the particles from closing properly. These findings provide a clearer picture of the mechanics behind protein assembly. They also show how related proteins can assemble into similar structures in different types of life, helping to trace their evolutionary history.

Accelerating research with Claude

Anthropic’s biology team used Claude throughout the entire investigation to speed up the research. The AI assisted with identifying promising proteins, designing DNA, and performing purification and microscopy analysis. One major breakthrough involved simplifying the particle isolation process, which traditionally takes about three days.

With Claude's help, the team developed a chromatography-based approach that reduced the purification time to roughly 90 minutes. The AI also helped identify the correct equipment and optimize the experimental procedure. Furthermore, the team used Claude to process cryo-EM data, producing their first complete structural model within 34 hours.

The researchers noted that conventional analysis for such a model could have required several weeks. However, the project also showed the AI's limitations. Claude produced an inaccurate map after skipping a step and suggested an additive that interfered with DNA delivery, both of which required human supervision to correct.

The next steps for vault research

While this structural mapping is a significant milestone, the central mystery of vault function remains. Scientists still do not know what these particles do inside living organisms. Leonard Rome, one of the original discoverers of vault particles, has described the question as enormously challenging.

The new findings provide a simpler way to investigate structures found in human cells by using bacterial models. This could allow for easier manipulation and observation of protein assemblies that were previously difficult to study. It opens a new door for researchers to test theories about cellular transport and defense.

The team shared their laboratory observations with the developers of Claude to improve the AI's performance in biology. This collaborative approach aims to refine how AI can assist in complex structural biology. Future research will likely focus on determining the specific biological roles of these mysterious barrel-shaped structures.