Scientists have just gotten their first clear look at vault particles in bacteria — and it could help crack one of cell biology’s longest-running puzzles. Anthropic’s biology research team used its AI model, Claude, to help find, purify, and map a vault-like particle built by a bacteriophage protein inside bacteria. It is the first 3D structure of a vault found outside complex (eukaryotic) cells.

This is a lab’s own report of its work. The paper has been submitted to the bioRxiv preprint server and has not yet been peer reviewed, so treat the details as early but promising.

What a vault particle actually is

A vault is a large, hollow, barrel-shaped protein shell. It was discovered in 1986 by Leonard Rome and Nancy Kedersha, who named it after the ribbed ceiling of a cathedral. Vaults show up in nearly every cell in your body and across most of the eukaryotic tree of life, from slime molds to humans.

Here is the strange part: after four decades, nobody knows what vaults do.

Researchers have proposed roles in moving molecules around the cell, shielding cells from toxic compounds, and supporting immune defense. None of those ideas has been confirmed. Mice bred without vaults still develop normally. Yet evolution kept this enormous structure around, which suggests it matters somehow.

Why finding vault particles in bacteria is a big deal

Part of the reason vaults stayed mysterious is that the lab workhorses biologists rely on — baker’s yeast, fruit flies, nematode worms — do not have them. That removes the easiest genetic tools from the table.

Bacteria are fast, cheap, and easy to engineer. A simpler bacterial system gives researchers a new handle on the problem.

Studying bacterial relatives of human proteins has paid off before. Bacterial versions of cGAS–STING signaling, gasdermins, and viperins each helped reveal the deep origins of human immune defenses.

The difference this time is that the human function is still unknown. The bacterial relatives may be the clue that unlocks it.

How the team built the structure

The work moved through three stages, with Claude assisting at each one under the researchers’ direction.

Searching

Claude Science searched a database of roughly 15 billion predicted proteins for vault relatives. They turned out to be rare but widespread: present in 29 bacterial phyla and four archaeal phyla, but in under 1% of species with sequenced genomes.

Building and isolating

The team picked 44 candidates and tested them in a cell-free system. The best performers moved into living E. coli, where 15 of 24 produced soluble protein. The conventional purification method takes about three days. By switching to size-exclusion chromatography — a route Claude initially argued against, then helped optimize — the team cut that to roughly 90 minutes from cells to microscope grid.

Under the electron microscope, 7 of those 15 proteins had assembled into vault-like particles.

Mapping

Cryo-electron microscopy confirmed the vault shape for four of the seven: two from phages and two from bacteria. The team chose a phage protein called L05 for the detailed structure.

The result: a shell built from 96 copies of a single protein, 48 per half, with a combined mass near 9.5 megadaltons. Human vaults use 78 copies. The phage vault is shorter and wider (about 54 by 45 nanometers versus 67 by 40 in mammals) yet encloses roughly 25% more volume because it is rounder. Its caps close in three tiered steps, like a stepped dome.

A first structural model was ready 34 hours into processing — work the team estimates would normally have taken weeks.

They also made it bigger

Working with Claude, the researchers made 15 altered versions of the protein. Some single changes near the cap stopped the shell from closing, producing open rosette-like particles instead. That pinpoints the region needed to seal the vault.

Then they copied two of the wall’s repeating segments, giving the protein nine repeats instead of seven — the same count as human vaults. It assembled into a larger vault, with about 20% more enclosed volume. That hints at how vaults may have changed size over evolutionary history.

What is still unknown

Plenty, and the team says so plainly.

• The function of vaults is still unsolved. This work deepens the mystery rather than settling it.
• A small molecule sits in a pocket in each of the 96 copies. It shows up clearly in the map, but nobody has identified it. Its shape resembles coenzyme A, which would be an early hint at a function — if it holds up.
• These vaults have only been made in E. coli so far. Whether they form in the bacteria and phages where the genes naturally live has not been tested.
• Suggested uses such as nanocapsules for delivering enzymes are speculative ideas, not demonstrated applications.
• The findings come from a company’s own research post about its own AI tool, and the preprint is not yet peer reviewed.

The team was also candid about the rough edges of working this way: one AI-written protocol called for an additive that interferes with getting DNA into cells and set the project back two weeks, and an early processing step produced a flawed map.

Why this is genuinely exciting

In under four months, a small team with limited structural biology experience screened dozens of candidates and solved the first vault structure found in prokaryotes. Most of the work, apart from the heavy cryo-EM computation, could be done by a typical lab.

That is the real headline. Vault particles in bacteria now give the whole field a faster, simpler system to study — and more people can run the experiments.

Four decades in, the question of what vaults do is still open. But for the first time in a long while, there is a new door to walk through.

Read the full write-up: Investigating vault particles with Claude