Claude mapped the first prokaryotic vault structure

Anthropic’s biology research team has mapped the first 3D structure of a vault particle assembled inside a prokaryotic cell, the simple single-celled kind such as bacteria, with Claude running the searches, designs and image processing under the team’s direction, the company said in a November 4 post. The time savings in the write-up are Anthropic’s own comparisons with how its lab normally works, and the team says it still has not determined what vaults actually do.

A vault is a large hollow protein shell present in nearly all the cells in our body, discovered in 1986 by Leonard Rome and Nancy Kedersha and without a confirmed function since. Until now vaults were known only in eukaryotes, the organisms with complex cells. Distant relatives of the vault protein had already been spotted in the genomes of bacteria and of the phages that infect them; what Anthropic says had not been shown before is that those proteins assemble into complete vault-like particles. The results are in a preprint submitted to bioRxiv.

What the phage vault looks like

The structure came from cryo-electron microscopy, where many copies of a particle are flash-frozen, imaged with an electron beam and combined into one 3D model. The particle is built from 96 copies, 48 per half, of a single phage protein the team calls L05, with a combined mass of about 9.5 megadaltons; the mammalian vault has 39 copies per half, 78 in total. It is shorter and wider than the human version, roughly 54 nanometers long and 45 wide against 67 by 40, and still encloses about 25% more volume because it is more spherical. Its wall is made of seven repeating segments stacked end to end rather than nine, and its end caps are flatter.

Those caps are the part the team singles out. A wall 48 copies around has to narrow toward the pole, and it does so in three steps: in every group of three neighbouring copies, one stops at the first step, one at the second, and one runs to the tip. The three are identical along the wall and take different shapes in the cap.

Working with Claude, the team made 15 altered versions of the protein, most with changes where neighbouring copies touch. Some single changes near the cap kept the shell from closing at all, leaving open, rosette-like particles instead of sealed vaults. Copying two of the wall segments to give the protein nine, the human number, produced larger vaults with about 20% more enclosed volume, which is one route by which vaults could have changed size over evolutionary time without breaking assembly.

Claude’s objection

The conventional way to isolate vaults, developed for eukaryotic ones, spins the sample through a tube of increasingly dense sugar and takes around three days. The team wanted to use size-exclusion chromatography instead, where the sample runs through a column of porous beads and the particles too big to enter the pores come out first. Claude argued against it, because clumps of misfolded protein leave the column in that same fraction. The team went ahead, and Claude then recommended a specific column and ways to track the vault protein and check whether particles had formed.

Claude also proposed skipping purification entirely and loading the contents of ruptured cells straight onto the column, which the team was reluctant to try because the columns are expensive and raw E. coli lysate can clog them. It worked. Anthropic is not the first group to isolate vaults on sizing columns, but says its version takes about 90 minutes from cells to microscopy grid rather than days.

For the structure itself, Claude worked in an internal multi-agent harness with access to internal compute, set up the open-source package RELION and ran each step while the team made the analysis calls. A first full model was ready in 34 hours and the final maps about a week after processing began, against the weeks or months Anthropic says the manual route takes it. That gap is an estimate against its own usual workflow, not a measured comparison, and it follows the same pattern as the four hours against about three months by hand Anthropic reported in October for a regulatory draft written for Enveda.

From 15 billion sequences to four structures

The search started in Claude Science across a database of about 15 billion predicted proteins. Vault relatives turned out to be rare but widely spread: present in 29 phyla of bacteria and four of archaea, but in less than 1% of species with a sequenced genome. The team picked 44 of the proteins and tested each in a cell-free system; Claude turned the chosen sequences into DNA for the host cell, added the control sequences that switch a gene on and prepared the designs for ordering. The best performers went into living E. coli, where 15 of 24 produced soluble protein, seven of those 15 formed vault-like particles under an electron microscope, and cryo-EM confirmed the vault shape for four of the seven, two from phages and two from bacteria. L05 was chosen for the detailed structure because it was among the best at forming complete, closed particles.

Rome, whom Anthropic brought on to advise the project, was shown the structure. “Figuring out the function of vaults is still enormously challenging,” he said, adding that the breadth of the findings “suggests that it is going to be a complicated function and an interesting ride.”

One pocket, one unknown

Claude flagged a feature in the map that the team cannot explain: a pocket partway up each of the 96 protein copies holds a small molecule that shows up clearly and has not been identified. Its shape resembles coenzyme A, a common helper molecule cells use to move chemical groups around. If it is coenzyme A or something close to it, that would be the first hint of what the shell is for. Anthropic says the structures and further data are to be released soon.