1. The Hook: What If Every Room in Your House Had 10,000 Locked Treasure Chests—and For 40 Years, Nobody Knew What Was Inside?

Imagine waking up tomorrow morning and discovering 10,000 tiny, hollow, barrel-shaped treasure chests stacked inside your bedroom.

You run next door to your neighbor’s house—they have 10,000 of the exact same barrels. You check your dog, a goldfish, and even a puddle of forest slime mold—every single one of them has thousands of these identical ribbed barrels inside! Yet when you ask the smartest scientists in the world what these barrels actually do, they shrug and say: “We found them 40 years ago, and we still have no idea.”

That is not a fairy tale. Right now, inside almost every single one of the trillions of cells that make up your body, there are roughly 10,000 giant, hollow protein barrels floating around (Kedersha & Rome, 1986, PMC2114306; Athukoralage et al., 2026). When cell biologists Nancy Kedersha and Leonard Rome first spotted them under an electron microscope at UCLA in 1986, the arched ribs of the shell reminded them of the curved stone ceilings (vaults) of a medieval cathedral. So they named them Vault Particles.

For 40 years, vault particles have been one of the biggest cold cases in biology. They are 1.8 to 3 times heavier than a ribosome (the molecular factory that builds every protein in your body), making vaults the largest known non-viral protein-and-RNA containers inside complex cells. Why would Mother Nature spend precious energy building 10,000 colossal cathedral barrels inside every human cell if they didn’t have a super-important job?

On November 4, 2026, a small biology team at the AI company Anthropic published a remarkable announcement (Investigating vault particles with Claude) alongside a 35-page primary technical manuscript (Athukoralage et al., 2026, Prokaryotic homologs of the major vault protein form engineerable vault nanocapsules). None of the researchers had ever worked with a protein barrel this huge before, and they had limited experience in mapping 3D atomic structures. Yet in under 4 months, teaming up with their AI assistant Claude, they:

  1. Searched 15 billion protein sequences across the tree of life and discovered that bacteria and bacteria-hunting viruses (bacteriophages) also build physical vault barrels!
  2. Cut the laboratory time needed to wash and isolate vault barrels from 3 days (4,320 minutes) down to 90 minutes—a 48× speedup.
  3. Built the first-ever 3D atomic map of a non-eukaryotic vault particle (from a virus called L05, resolved down to 3.2 Å, or 0.32 nanometers), revealing a wider, 96-piece barrel that holds ~25% more cargo space than a human vault—and then engineered a “stretch limousine” version (L05+2) that holds ~50% more space than a human vault!

Wait—did Claude magically solve the 40-year mystery by itself? No! When we open the primary laboratory receipts, we find a thrilling detective story where humans had to overrule Claude’s cautious warnings, Claude later invented a daring lab shortcut that scared the humans, and Claude made two huge mistakes—including a chemical recipe blunder that wasted two full weeks and a math shortcut that accidentally blended a 3D molecule with its own mirror image!

Grab your magnifying glass. Let’s start from Step Zero and see how these microscopic cathedrals work.

2. Starting from Zero: What Is a Vault Particle, and Why Was This Case Stuck for 40 Years?

To understand why biologists were stuck for four decades, we first need to know the two main types of living cells on Earth and how microscopic building blocks snap together.

Step 1: Complex Mansion Cells vs. Tiny Studio-Apartment Cells

Every living thing on Earth falls into one of two giant neighborhoods:

  • Complex Cells (Scientists call these Eukaryotes): Think of a human cell, a mouse cell, or a tree cell like a giant two-story mansion with separate rooms having locked doors—a bedroom for the DNA blueprints (the nucleus), a power plant (mitochondria), and hallways full of delivery trucks.
  • Simple Cells (Scientists call these Prokaryotes): Think of bacteria (like E. coli in your gut) and their ancient cousins (Archaea) like a tiny, one-room studio apartment. There are no inner walls; the DNA blueprints, tools, and food all float together in one single room. And floating around outside those studio apartments are microscopic robot pirates called Bacteriophages (or Phages for short)—viruses that land on bacteria and hijack their machinery to make more viruses.

Step 2: How 78 Identical LEGO Strips Snap Into a Hollow Cathedral Barrel

Inside your cells, proteins are like custom-molded plastic LEGO pieces. Most proteins are tiny tools—a pair of scissors here, a hinge there.

A Vault Particle is completely different. In humans, the cell prints 78 identical copies of one long ribbon-shaped protein.

  • 39 ribbons stand side-by-side in a circle, zipping their edges together like the wooden slats of a pickle barrel to form a bottom half-cup.
  • Another 39 ribbons zip together to form an identical top half-cup.
  • The two half-cups snap together waist-to-waist to form a sealed, hollow barrel 67 nanometers (nm) tall and 40 nm wide (Tanaka et al., 2009; Athukoralage et al., 2026).
  • Scale Check: How tiny is 67 nanometers? A single strand of hair on your head is about 75,000 nanometers wide. That means you could park more than 1,100 human vault barrels end-to-end across the width of one single human hair! Yet inside the microscopic world of a cell, a 67 nm barrel weighing 13 megadaltons (MDa) (about the weight of 13 million hydrogen atoms) is a towering skyscraper—nearly twice as big as the cell’s protein factory (ribosome).

Scientists call the single protein ribbon that builds this shell the Major Vault Protein (MVP).

Step 3: Why Couldn’t Scientists Figure Out What Vaults Do?

Usually, when biologists want to know what a cell part does, they play a simple game called “Break It and See What Goes Wrong.” It’s like unplugging one mystery wire under the hood of a car: if the headlights go dark, you know that wire controls the headlights!

For 40 years, that game failed on vaults for two infuriating reasons:

  1. The “Lab Pets” Lost Their Vaults: Scientists do 90% of their fast genetic experiments in four easy-to-grow “lab pets”: baker’s yeast, fruit flies, tiny roundworms (C. elegans), and mustard plants. By sheer bad luck, all four of those creatures lost their vault genes millions of years ago (Daly et al., 2013, PMC3762200)!
  2. Vault-Free Mice Act Totally Normal: Mice do have vaults. So in the early 2000s (and again in 2024), scientists bred special mice with broken vault genes (Liu et al., 2004, PMC419869; Prajapat et al., 2024, PMC11482699). Everyone expected the mice to get sick. Instead, the vault-free mice ran around their cages eating snacks, growing up healthy, and having babies! That told scientists that vaults aren’t like the car’s engine (which you need every second); they are more like an airbag or a fire extinguisher—you only notice it’s missing when a specific disaster strikes!

Step 4: The Smart Backdoor—Look in Bacteria!

Wait: if studying vaults in slow-growing mice is too hard, what if simpler bacteria or viruses (phages) also have vaults? Bacteria grow in 20 minutes in a warm flask of soup! Better yet, over the last seven years, scientists discovered that several famous human immune shields—like cGAS-STING (Cohen et al., 2019, Nature) and viperins (Bernheim et al., 2021, Nature)—actually started out billions of years ago inside bacteria as shields to fight off invading phage viruses!

Computer scientists had already spotted MVP-like genes in bacterial and viral DNA databases (Daly et al., 2013, PMC3762200). But nobody knew if those bacterial genes actually built real, 3D hollow barrels, or if they were just flat membrane patches.

That is the exact mystery Anthropic’s scientists and Claude set out to solve.

flowchart TD
    subgraph SG1["The 40-Year Roadblock in Complex Cells (Eukaryotes)"]
        E1["Human & Mouse Cells Have ~10,000 Vaults Each<br/>(78 MVP Chains, 67 nm Tall x 40 nm Wide)"]
        E2["Missing in Yeast, Flies & Worms;<br/>Knockout Mice Look Fine in Clean Cages"]
        E3["3-Day Sugar-Tube Spin Needed to Isolate Vaults"]
        E1 --> E2 --> E3
    end

    subgraph SG2["Anthropic + Claude's Bacterial & Phage Shortcut"]
        P1["Step 1: Search 15 Billion Proteins with Claude<br/>(Expand Known MVP Groups from 41 to 1,507)"]
        P2["Step 2: Test 44 Designs in Test Tubes & E. coli<br/>(15 Make Soluble Protein -> 7 Form Barrels)"]
        P3["Step 3: 90-Minute Bead-Column Shortcut (SEC)<br/>(48x Faster Than the Old 3-Day Spin!)"]
        P4["Step 4: Flash-Freeze & Map L05 Phage Vault in 34 Hours<br/>(96 Chains, 3.2 Å Atomic Map + Engineered Mega-Vault)"]
        P1 --> P2 --> P3 --> P4
    end

    SG1 -->|"Shift to Fast Bacterial Lab"| SG2

3. The Secret Decoder Ring: Grown-Up Words Translated

Before we open the laboratory notebooks, let’s translate the seven fancy scientific terms you’ll see in the report into plain 5th-grade English.

The Grown-Up Word / Acronym What It Actually Means in Plain English Everyday 5th-Grade Analogy
Vault Particle A giant, hollow, barrel-shaped container inside cells built from dozens of identical protein ribbons zipped together. A hollow plastic Easter egg made of 78 or 96 interlocking LEGO strips that snaps together at the middle belt.
Major Vault Protein (MVP) The single main protein building-block ribbon that repeats side-by-side to form the entire outer shell of a vault. One curved wooden plank used to build a pirate’s wooden barrel.
Prokaryote vs. Eukaryote Prokaryotes are simple one-room cells without a nucleus (bacteria and archaea). Eukaryotes are complex cells with many inner rooms (humans, mice, plants). A tiny one-room camping tent (prokaryote) vs. a two-story house with bedrooms and hallways (eukaryote).
Bacteriophage (Phage) A microscopic virus that infects bacteria (L05 is a vault protein carried by a phage). A tiny lunar lander robot that lands on a bacterium and hijacks its 3D printers.
Size-Exclusion Chromatography (SEC) A fast way to separate giant particles from tiny molecules by pouring liquid through a tube packed with microscopic sponge beads that have tiny tunnels inside. Rolling giant basketballs and tiny marbles down a hallway full of mouseholes: the marbles get detoured into the mouseholes, while the giant basketballs roll straight out the exit first (excluded/void volume)!
Cryo-Electron Microscopy (Cryo-EM) Flash-freezing thousands of protein barrels in glass-like ice and taking shadow pictures with an electron beam from every angle to build a 3D atomic map. Taking 34,000 X-ray photos of a falling snowflake from every side and using a computer to stitch them into a 3D statue.
Angstrom (Å) & Nanometer (nm) Ultra-tiny rulers for measuring atoms and molecules. 1 nm = 10 Å. A single carbon atom is about 1.5 Å wide. If a single human hair (75,000 nm) were as wide as a football stadium, 1 nanometer would be the width of a blade of grass, and 3.2 Å would be the thickness of a dime!

4. The Big Claim vs. The Detective Check: Opening the Lab Receipts

How did a small team with limited structural biology experience go from 15 billion database sequences to the world’s first 3D prokaryotic vault map in less than four months? Let’s check the exact numbers in their primary technical report (Athukoralage et al., 2026).

Receipt #1: The 15-Billion-Protein Treasure Hunt (36.8× More Vault Cousins)

First, the scientists asked Claude to help build 14 mathematical search nets (called Profile Hidden Markov Models, or pHMMs) trained to recognize the “shoulder” joint of known vault proteins.

  • They cast those nets across a gigantic global database of ~15 billion predicted proteins (grouped into 1.94 billion clusters).
  • Before this study, scientists only knew about 41 distinct groups (at 50% sequence similarity) of MVP and MVP-like proteins. Anthropic’s search uncovered 1,507 groups—expanding the known vault family by 36.8×!
  • When they checked 111,922 different species of bacteria and archaea, they found vault genes in 29 bacterial phyla and 4 archaeal phyla, plus inside bacteriophages (viruses).
  • The Rarity Clue: Even though vault genes show up across almost every branch of the bacterial family tree, fewer than 1% (565 out of 111,922, or ~0.5%) of bacterial species carry one! Just like a specialized snowplow that only a few towns keep in their garage, bacteria appear to pass vault genes back and forth (often via phage viruses) for a special job.

Receipt #2: The 90-Minute Basketball-and-Marble Trick (Where Humans Beat Claude, Then Claude Beat Tradition!)

From those 1,507 groups, the team picked 44 diverse vault genes to test in the lab. Claude rewrote their genetic letters (codon optimization) so laboratory E. coli bacteria could read them easily, and added genetic “ON/OFF” switches (T7 promoter and terminator).

  • In quick 4-hour test-tube trials (cell-free expression), 24 candidates looked promising.
  • Inside living E. coli bacteria, 15 out of 24 (62.5%) produced dissolved, usable protein.

Now came the bottleneck: how do you fish intact vault barrels out of popped bacterial soup?

  • The Old Way (3 Days): For 40 years, scientists spun cell soup in an ultra-fast centrifuge through layers of thick sugar syrup for ~72 hours (3 days). Testing 15 proteins that way would take weeks of tedious spinning.
  • The New Idea (Size-Exclusion Chromatography, or SEC): Remember our basketballs-and-marbles hallway from the Decoder Ring! A tube packed with microscopic porous sponge beads (Superose 6 Increase) traps tiny proteins (< 5 MDa) inside the bead tunnels, while giant 9.5 MDa vault barrels are too huge to fit in the holes—so they wash straight out in the first 12 minutes (between 7.97 mL and 9.72 mL, called the excluded volume or void volume).
  • Wait—Look Who Said What:
    1. First, Claude argued AGAINST using the bead column! Why? Because Claude knew from textbooks that ugly, tangled balls of broken protein trash (aggregates) are also huge, so they wash out in the exact same first peak as real vaults! The human scientists overruled Claude and decided to try it anyway—checking the first peak under an electron microscope to tell real barrels apart from trash clumps.
    2. Second, Claude proposed a daring shortcut that made the humans nervous! Normally, before you pour popped bacterial soup onto an expensive glass chromatography column, you spend hours pre-cleaning it so bacterial gunk doesn’t clog the column. Claude suggested: “Why not squirt freshly popped, clarified E. coli soup straight onto the column with zero pre-purification?” The humans hesitated because columns cost thousands of dollars—but they tried Claude’s daring idea, and it worked cleanly, shrinking vault isolation from 3 days down to 90 minutes from living cells to microscope grid!

Out of the 15 proteins tested with this 90-minute method, 7 formed barrel-like structures under a quick negative-stain microscope, and 4 were confirmed as true, closed vaults using high-powered cryo-EM (L05 and C0 from phage viruses, plus Hmin and Bc1 from bacteria).

Receipt #3: Inside the L05 Phage Vault—96 Ribbons, an Origami Roof, and a Stretch-Limo Upgrade!

The team zoomed in on L05, a vault protein from an uncultured bacteriophage virus. Using 3,041 cryo-EM microscope movies (11,675 pristine closed vault particles), Claude ran the open-source RELION 5.0.1 software across hundreds of GPUs in parallel, delivering a first 3D atomic model in just 34 hours at a crystal-clear 3.2 Å resolution (2.95 Å at the waist)!

Look at how the viral L05 Vault and the engineered L05+2 Mega-Vault compare to the Human Vault across every measurement in the technical report:

Measurement / Feature Human Vault (Mammalian MVP) Phage L05 Vault (Wild-Type) Engineered L05+2 Mega-Vault Plain-English Detective Takeaway
Natural Owner Humans & complex cells (Eukaryotes) Bacteriophage virus (infects bacteria) Engineered in lab by duplicating wall parts First experimental proof that non-eukaryotes build 3D vaults!
Total Protein Ribbons (Chains) 78 chains (39 top + 39 bottom) 96 chains (48 top + 48 bottom) 96 chains (48 top + 48 bottom) Phage vault packs 18 more ribbons around its circle.
Stacked Wall Rings (Repeats) 9 repeats per chain (35.3 nm wall) 7 repeats per chain (28.1 nm wall) 9 repeats per chain (repeats 4 & 5 copied) Copying 2 wall segments mimics how evolution stretches barrels.
Outer Length × Width 67 nm × 40 nm (Tall lemon shape) 54.5 nm × 47.5 nm (Plump egg shape) 58.5 nm × 49.4 nm (Giant egg shape) L05 is 12.5 nm shorter but 7.5 nm wider than human vaults.
Inside Cargo Space (Volume) Baseline (100%) ~125% (+25% larger) ~150% (+50% larger than human!) A rounder egg holds more room inside than a skinny lemon!
Waist Glue Area (per chain) ~420 Ų ~225 Ų (0 shared amino acids) Rotated by ~3° at the waist seam L05 uses half the waist glue area with totally different letters.
Mystery Cargo Pocket Inside Binds NAD+ & ADP nucleotides Mystery blob matching Coenzyme A Same Coenzyme A-like blob preserved Both human and viral vaults hold cellular fuel-shuttle molecules!

Three Cool Engineering Secrets Inside L05:

  1. The “Rule of 3” Origami Dome Roof (48 → 16 → 8 Symmetry): Imagine trying to sew 48 identical fabric strips together to close the top of a round tent. As all 48 strips march toward the north pole, they run out of room and jam into each other! How does L05 solve this traffic jam without needing a second protein? Every group of 3 neighboring identical chains (A, B, C) plays a different game of origami at the roof:
    • Chain A stops halfway up and folds inward into the barrel (Step 1).
    • Chain B marches higher and folds outward toward the outside (Step 2).
    • Chain C (only 16 of the 48 chains!) marches all the way to the very center tip to seal the hole—where chains C and C' take turns again (8-fold symmetry, Step 3)!
  2. One Single Letter Flip Pops the Barrel Open into a Flower (Rosette): To prove which Velcro strips hold the barrel shut, the team built 15 mutated versions of L05. Flipping just one single amino-acid letter on the wall seam (R55E, turning a positive magnet into a negative magnet so neighbors repel) or in the cap (I794D) completely stopped the barrel from closing—turning sealed barrels into flat, open flower shapes called rosettes!
  3. The Hidden Clue in Pocket #1 (Coenzyme A): When Claude inspected the 3D cryo-EM map, it noticed an extra, unexplained blob sitting inside a tiny pocket on the inner wall (lumen) of all 96 chains (between repeats 1 and 2). Claude tested 25,000 known chemical shapes against that mystery blob and found the best fit was Coenzyme A (CoA)—a universal cellular battery-pack helper that carries chemical building blocks inside every living cell!
flowchart LR
    subgraph SG1["The Funnel: From 15 Billion Sequences to 1 Atomic Structure"]
        F1["15,000,000,000<br/>Database Proteins"] -->|"14 Shoulder HMMs"| F2["1,507 MVP Groups<br/>(Up from 41!)"]
        F2 -->|"Pick Diverse Set"| F3["44 DNA Designs<br/>(Test-Tube Screen)"]
        F3 -->|"Move Best to E. coli"| F4["24 Tested in Cells<br/>(15 Soluble Proteins)"]
        F4 -->|"90-Min SEC Column"| F5["7 Form Barrels<br/>(4 Confirmed by Cryo-EM)"]
        F5 -->|"34-Hr GPU Run"| F6["L05 Phage Vault<br/>(3.2 Å Atomic Map)"]
    end

5. The Four Suspects: What Actually Caused This Breakthrough (and Why Do Bacteria Have Vaults)?

Following our critical-thinking detective rulebook, whenever someone says “Factor X caused Outcome Y,” we must interrogate all Four Suspects (Direct Cause, Reverse Cause, Hidden Third Factor, and Feedback Loop). Here we have two linked mysteries:

  • Biological Mystery: Why do bacteria and phage viruses carry vault barrels?
  • AI Speed Mystery: Did Claude really make structural biology 10× faster, or was something else doing the heavy lifting?
Suspect Type The Biological Suspect (Why Phages & Bacteria Have Vaults) The AI Speed Suspect (Why the Project Finished in <4 Months) Detective Verdict & Primary-Source Proof
Suspect #1: Direct Cause (X → Y) Vaults are specialized cargo/enzyme cages that trap helper fuels (Coenzyme A or NAD+) and partner enzymes to run chemical defenses or repairs inside a sealed room. Claude’s end-to-end automation (writing all bioinformatics code, designing 44 DNA constructs, suggesting raw-lysate SEC, and scripting RELION) directly compressed months of coding and lab setup into days. Supported (With Guardrails): Gene-neighborhood scans of 2,185 bacterial vault loci show they sit right next to anti-phage immune genes (JAB), RNA repair (RtcB), and nucleosidases. And Claude wrote 100% of the code.
Suspect #2: Reverse Cause (Y → X) Rather than bacteria inventing vaults for themselves and phages stealing them, giant phages (bacteriophages) may have evolved or modified vault shells as viral storage compartments and spread them into bacterial hosts during infections. Rather than Claude making the science easy, picking a naturally self-assembling homopolymer (96 identical copies of 1 protein) made the wet-lab and symmetry math ideal for an AI coding harness to succeed! Strong Contributor: Because vaults assemble from 1 single gene (MVP alone) and weigh 9.5 MDa, they are uniquely suited to 1-step SEC isolation and high-symmetry (D16/C48) cryo-EM averaging.
Suspect #3: Hidden Third Factor (Confounder Z) Both bacterial vaults and eukaryotic vaults evolved from a common ancient ancestor: membrane-anchored SPFH ring proteins (like flotillins and stomatins) that pinched off the cell wall to become free-floating barrels. Massive Industrial Compute & Pre-Built Open-Source Tools: Running RELION 5, AlphaFold2/OpenFold3, and DIAMOND across hundreds of GPUs in parallel, plus contracted Thermo Fisher Krios G4 microscope time, drove the 34-hour turnaround. Crucial Confounder: Without hundreds of parallel GPUs and gold-standard open-source software (RELION, MAFFT, MMseqs2), Claude’s Python scripts could never have solved a 3.2 Å map in 34 hours.
Suspect #4: Feedback Loop (X ↔ Y) Host-Virus Arms Race: Bacteria use vault cages for metabolic/immune defense (RtcB RNA repair), phages hijack and resize those barrels (adding/removing wall repeats) to counter-attack, and hosts adapt again. Human-AI Co-Correction Loop: Humans overruled Claude’s fear of SEC; Claude invented the raw-lysate shortcut; humans caught Claude’s mirror-image map; and the team fed lab notes back to Anthropic ML trainers to upgrade newer Claude models! The Real Winner: Neither humans alone nor Claude alone could have moved this fast. The breakthrough came from the tight loop between human domain judgment and AI execution speed.

6. The Catch: Where Did Claude Mess Up, and What Is Still Unsolved?

Headlines on social media love to pretend that AI scientists are flawless wizards. To Anthropic’s credit, their report is refreshingly honest about six big catches where Claude stumbled or where the science remains unfinished:

Catch #1: The 2-Week Poisoned-Soup Blunder

When the team asked Claude to write a laboratory recipe (protocol) for growing bacteria and putting custom DNA inside them (transformation), Claude included a chemical additive in the growth broth that blocks DNA from entering bacterial cells! Because the human scientists weren’t experts in that specific step, they trusted the recipe—and lost two full weeks before figuring out why their experiment failed.

Catch #2: The Hallucinated Mirror-Image Map (“Please Let’s Do Normal Things!”)

In cryo-EM, 2D shadow pictures cannot easily tell a left-handed corkscrew apart from a right-handed corkscrew (chirality / handedness). Early in the image processing, instead of following the standard textbook step, Claude improvised a shortcut that blended the 3D vault map with its own mirror image, creating a scrambled, physically impossible protein! The human scientists had to step in and literally type to Claude:

“Please let’s do normal things unless they don’t work.”

Catch #3: Why Claude’s Initial Warning About the Bead Column Still Matters

Remember how Claude initially warned against using the SEC sponge-bead column because broken protein trash also comes out in the fast excluded volume peak? Look at Figure 4 in the technical paper (Athukoralage et al., 2026, p. 11): when the team made the 5 broken mutant proteins (R55E, I794D, etc.), all 5 broken mutants still came out in the exact same fast SEC peak as real vaults, even though the microscope showed they were just broken flower rosettes or messy clumps! And one protein (Bc1) looked like flat rings under negative-stain microscopy but turned out to be full barrels in cryo-EM. Lesson: A fast bead column never replaces looking through an actual electron microscope!

Catch #4: The “Hundreds of GPUs” Supercomputer Catch (Plus Fitted Noise in L05+2)

Could a high school or small college biology lab repeat the 90-minute bead-column purification tomorrow? Yes! That part only uses standard lab tubes and a column. Wait—could a small college lab solve the 3D cryo-EM structure in 34 hours? No! Anthropic admits that hitting 34 hours required running hundreds of graphics chips (GPUs) in parallel inside Anthropic’s internal cloud, plus collecting 3,041 movies (and 16,824 movies for L05+2) on an elite 300 kV Thermo Fisher Krios G4 electron microscope. Furthermore, in the Methods section (page 34), the authors note that pushing the L05+2 mega-vault refinement to convergence increased half-map differences (1.09× to 1.89× in the wall), indicating some fitted noise (overfitting).

Catch #5: Factory E. coli Is Not Wild Nature!

Every single vault barrel in this study was produced artificially by forcing laboratory E. coli bacteria (BL21(DE3)) to crank out huge amounts of protein from a synthetic gene switch (pET-21(+) with 0.5 mM IPTG). As the authors openly state:

  • They have not yet tested whether L05 or bacterial MVP proteins actually assemble into vaults inside wild bacteria or real phage infections in nature.
  • Because the vaults were built inside E. coli rather than their native host, their natural partner proteins and true biological cargo were left behind—which is why the 40-year mystery of what vaults actually do for a living cell is STILL unsolved!

Catch #6: Commercial & Patent Disclosures

Following our primary-source conflict-of-interest rule: all authors are employees of Anthropic (which sells Claude and Claude Science), Dr. Leonard Rome (co-discoverer of vaults and co-founder of Vault Pharma) was a paid consultant on the manuscript, and Anthropic has filed a provisional patent application covering these engineered prokaryotic vault nanocapsules.

7. The Final Report Card: What We Actually Think

Even though the 40-year mystery of the vault’s natural job remains open, proving that bacteria and phages build customizable, 96-chain vault barrels—and showing that you can stretch their cargo space by +20% to +50% simply by copying and pasting wall-repeat blocks in E. coli with a 90-minute purification—is a genuine milestone for both evolutionary biology and bioengineering.

The “If … Then …” Decision Guide

  • If you are a biologist hunting for the 40-year secret of what vaults do, then stop waiting years for mouse experiments to show a symptom: take the bacterial vault species identified here (like the myxobacterium Hyalangium minutum, Hmin, where the vault gene sits next to immune defense, RtcB RNA repair, or nucleosidase partners), knock out the vault gene in its native bacterium, and attack that bacterium with phages and toxins to see what breaks!
  • If you are a bioengineer designing microscopic delivery trucks (nanocapsules) for enzymes or medicines, then prokaryotic vaults (L05 and L05+2) solve two huge headaches of human vaults: they can be grown cheaply inside standard E. coli bacteria (instead of finicky insect or mammalian cells) and isolated in 90 minutes instead of 3 days, with dial-a-size cargo volume (+25% to +50% larger than human vaults).
  • If you are a scientist using AI agents (Claude) in a real wet lab, then use AI aggressively to search billions of sequences, write RELION processing scripts, and challenge slow 40-year-old lab habits—but never trust an AI lab recipe or 3D map blindly in an unfamiliar domain without checking every chemical additive and verifying mirror-image handedness against textbook controls!

Probability-Weighted Scenarios (Next 3–5 Years)

Scenario Probability What Happens Next What Evidence Would Prove This Right
Base Case: Bacterial Genetics Cracks the Function & E. coli Vaults Become Lab Workhorses ~60% Academic labs use the 1,507 new MVP clusters and fast bacterial genetics to prove that prokaryotic vaults act as metabolic/stress cages (compartmentalizing Coenzyme A / NAD+ reactions or RNA repair during viral infection), while bioengineers adopt L05 as a cheap E. coli nanocapsule scaffold. Native-host knockout studies in Hyalangium minutum or phage infection assays showing clear survival defects when MVP or its neighboring operon partner (RtcB/JAB) is deleted; mass-spectrometry confirmation of the Coenzyme A pocket ligand.
Bull Case: Programmable “Dial-a-Size” Nanocapsules Reach Preclinical Drug & Enzyme Delivery ~25% By engineering the unresolved ~80-amino-acid inner loop (between repeats 6 and 7) as a universal “Velcro hook” for loading enzymes and mRNA, size-tuned prokaryotic vaults (L05+2, L05+4) outperform mammalian vaults in manufacturing yield and cargo capacity. High-yield cargo encapsulation papers demonstrating stable enzyme/mRNA loading inside L05+2 without triggering destructive immune reactions or aggregation in animal models.
Bear Case: Expression Artifacts & Human Immune Rejection Limit Biotech Use ~15% Several prokaryotic MVP homologs turn out to form closed barrels only when artificially overexpressed in E. coli (remaining flat membrane-associated SPFH-like oligomers at natural levels), and viral L05 protein shells prove too foreign (immunogenic) to inject into humans as drug carriers. Cryo-electron tomography (cryo-ET) of native bacteria/phage infections failing to find closed barrels in situ, or strong anti-L05 antibody neutralization in mammalian delivery tests.

Primary Sources & Verified References

  1. Anthropic Research Announcement (Nov 4, 2026): Investigating vault particles with Claude
  2. Primary Technical Preprint Manuscript (Nov 4, 2026): Januka S. Athukoralage, Peter H. Yoon, Amir Shanehsazzadeh, Eric Kauderer-Abrams, Nicholas T. Perry, & Matthew G. Durrant, Prokaryotic homologs of the major vault protein form engineerable vault nanocapsules (Anthropic Technical Report / bioRxiv Preprint PDF).
  3. Original 1986 Discovery of Vault Particles: Nancy L. Kedersha & Leonard H. Rome, Isolation and characterization of a novel ribonucleoprotein particle: large structures contain a single species of small RNA, Journal of Cell Biology 103(3), 699–709 (1986).
  4. Mammalian Vault Structure & Cap Architecture: H. Tanaka et al., Science (2009) / PMC3184113; F. Lapenta et al., Structural flexibility of the human vault particle revealed by high-resolution cryo-EM and molecular dynamics simulations, Nature Communications 17, 6033 (2026); K. Geißler et al., The vault associates with membranes in situ, Nature Communications 17, 3659 (2026).
  5. Human Vault NAD+ Binding & Partner Proteins: J. E. Lodwick et al., Structural insights into the roles of PARP4 and NAD+ binding in the human vault cage, Nature Communications 16, 6724 (2025).
  6. Knockout Mouse Viability Studies: Y. Liu et al., Vault poly(ADP-ribose) polymerase is associated with mammalian telomerase and is dispensable for telomerase function and vault structure in vivo, Molecular and Cellular Biology 24(12), 5314–5323 (2004); M. Prajapat, L. Sala, & J. A. Vidigal, The small noncoding RNA Vaultrc5 is dispensable to mouse development, RNA 30(11), 1465–1476 (2024).
  7. Evolutionary History & Model Organism Loss of Vaults: T. K. Daly, A. J. Sutherland-Smith, & D. Penny, In silico resurrection of the major vault protein suggests it is ancestral in modern eukaryotes, Genome Biology and Evolution 5(8), 1567–1583 (2013).
  8. Engineered Vault Nanocapsules: V. A. Kickhoefer et al., Engineering of vault nanocapsules with enzymatic and fluorescent properties, PNAS 102(12), 4348–4352 (2005).
  9. Bacterial Origins of Human Immune Systems: D. Cohen et al., Cyclic GMP–AMP signalling protects bacteria against viral infection, Nature 574, 691–695 (2019); A. Bernheim et al., Prokaryotic viperins produce diverse antiviral molecules, Nature 589, 120–124 (2021).