CRISPR · Bacteriophages · Metagenomics · Microbiology · Research BriefPaper of 29 October 2026
A census of bacterial immune memory
794 million CRISPR spacers, and what they say about phages
Bacteria and archaea keep a record of the viruses and plasmids they have met. Short fragments of foreign DNA, called spacers, are stored in CRISPR arrays and used to recognise the same invader later. On 29 October 2026 a team led by the DOE Joint Genome Institute published the largest survey of those records so far in Nature Microbiology. It mined 467,018 public metagenomes for spacers and matched them against databases of viruses and plasmids. The result: memories are highly variable and mostly short-lived, the ones that persist point at real virus–host pairs, and one group of phages seems to get into cells it cannot infect.
What Happened
A database of spacers from short reads Figures as the paper reports them
- 794MCRISPR spacers collected, 409.6 million of them distinct
- 367,609Public metagenomes with spacers, of 467,018 searched
- 1.18BHits against known viruses, from 38 million distinct spacers
- 67.0%Spacers seen on a single read
Confirmed
The paper is peer-reviewed and open access. The database is public at spacers.jgi.doe.gov, and the extraction tool, SpacerExtractor, is released under the GNU GPL on the institute’s GitLab and on bioconda.
Inference, by the authors
That non-host cells take up phage genomes through diversity-generating retroelements is a hypothesis drawn from correlations, not something the data observe directly.
Timeline
From scattered studies to one database Publication dates of each source
- Jul 2017Deep sequencing of cyanobacterial CRISPR arrays finds spacer diversity that does not saturate even at millions of sequences (Silas et al., mBio).
- Jan 2023IMG/VR v4 holds over 15 million virus genomes and fragments, in 8.7 million vOTUs (Camargo et al., Nucleic Acids Research).
- Apr–Nov 2023A landfill study and a study of phages of CPR bacteria report viruses targeted by spacers from distantly related hosts (George and Hug, Scientific Reports; Liu et al., mBio).
- Feb 2024The SpacerExtractor project is created on the institute’s GitLab (SpacerExtractor).
- Dec 2024A human gut study reports spacer acquisition as rare, about one spacer every 2.7 years in time series (Zhang et al., Cell Genomics).
- 22 Dec 2025The global spacer manuscript is received by Nature Microbiology (Roux et al.).
- 29 Oct 2026Published online, open access, after acceptance on 11 August 2026 (Roux et al.).
The Argument
Persistent spacers track real hosts; one kind of phage breaks the rule The authors’ claim, and what complicates it
Roux and colleagues argue that cells acquire spacers often, but most are lost to drift. The few that become common in a population are kept because they work. Common spacers matched known viruses or plasmids three times as often as rare ones, 30.5% against 10.1%. A virus targeted many times by one host lineage was usually targeted by its known host: 76.8% of pairs with 10 or more hits were consistent with it. Known hosts mostly left distant, mismatched hits, the trace of phages mutating to escape. The exception is viruses that carry a diversity-generating retroelement (DGR), which can retarget a phage’s tail fibres. They made up 57.6% of pairs with many near-exact hits from a non-host, against 16.8% of all pairs. The authors suggest such phages attach to and enter cells they cannot infect. The cell then records them, and since infection fails anyway, there is no pressure for the phage to escape.
- 57.6%Non-host, near-exact pairs involving a DGR virus, against 16.8% of all pairs
- 91.5%Well-targeted viruses hit by a single microbial class
What complicates it
The sampling is uneven: the authors note that Western-diet human gut samples are heavily over-represented. Human-associated spacers hit known viruses more often, 21.8% against 5.64%, which the authors put down mostly to databases favouring human microbiome viruses. Most repeats, 538,406 of 581,889, were found only in metagenome assemblies and could not be tied to a CRISPR type or a taxon. And a spacer hit shows only that a genome entered a cell, not that the virus could complete an infection there.
What Others Add
Earlier studies, and one that disagrees Each item as the named source gives it
| Question | What the record shows | Source |
|---|---|---|
| How often do gut bacteria acquire spacers? | Rarely, by one measure: about one spacer every 2.7 years in metagenome time series, and one every 2.9 point mutations in isolate genomes. In Bifidobacterium longum, a whole CRISPR system spreading between strains brought three times as many spacers as local acquisition. Roux and colleagues read their contrasting result as a difference of method: many spacers acquired by individuals, few becoming common. | Zhang et al., Cell Genomics |
| Is high spacer diversity real? | Targeted sequencing of arrays in a cultivated cyanobacterium found spacer diversity with no saturation even at millions of sequences, and few spacers with any known match. | Silas et al., mBio |
| What were spacers matched against? | IMG/VR v4: over 15 million virus genomes and fragments from genomes, metagenomes and metatranscriptomes, in 8.7 million vOTUs, 231,408 with a high-quality representative. | Camargo et al., Nucleic Acids Research |
| Where did cross-phylum targeting come from? | A two-year landfill study found four viruses predicted to infect across phyla, while most interactions still fitted one virus, one host. A study of phages of CPR bacteria found some targeted by spacers from both CPR and non-CPR bacteria. | George and Hug, Scientific Reports; Liu et al., mBio |
Conclusion
A map of immune memory, with its blind spots marked What the paper settles, and what it leaves open
Settled
Spacer collections are vast and mostly made of rare, single-sample memories. Spacers that persist match known viruses more often. Truly broad targeting is uncommon: only 1.65% of well-targeted viruses had no dominant host class.
Open
Whether DGR phages really enter non-host cells needs experiments. How fast spacers are acquired in the gut is disputed. Short reads cannot rebuild individual arrays, which the authors say long-read sequencing could.
- Do not infer host range from one spacer hit. Pairs with fewer than 10 hits disagreed with the known host 54.1% of the time.
- Look at the mismatch profile. Mostly distant hits suggest a real host under escape pressure; mostly near-exact hits from a non-host suggest entry without infection.
Quick notes
- Simon Roux et al., “Planetary-scale metagenomic analysis reveals CRISPR targeting dynamics across ecosystems”, Nature Microbiology, 29 Oct 2026
- DOE Joint Genome Institute, SpacerDB documentation
- Simon Roux, SpacerExtractor, GitLab
- An-Ni Zhang et al., Cell Genomics, 23 Dec 2024
- Sukrit Silas et al., mBio, 11 Jul 2017
- Antonio Pedro Camargo et al., Nucleic Acids Research, 6 Jan 2023
- Nikhil A. George and Laura A. Hug, Scientific Reports, 5 Apr 2023
- Jett Liu et al., mBio, 27 Nov 2023. All read 30 Oct 2026.
Reliable
Worth believing
Sharing a subject
No other article shares one yet.
Sharing a source domain
- Sequencing a Genome at Home — The Full ONT MinION Workflownature.com
- The Rise of Biomedical Foundation Models — 220 Models Surveyednature.com
- The Electron, to Thirteen Digitsnature.com
- Three Cuts Into the Uncuttablenature.com
- Not a Ladder, a Palimpsestnature.com
- Decades Rather Than Yearsnature.com
- Small Variants in the Long-Read Eranature.com