CRISPR was not invented first as a medical tool. It was discovered as a bacterial immune system: bacteria store genetic memories of viruses called phages and use Cas proteins as guided molecular scissors. MIT’s new work follows the same logic: if CRISPR came from bacterial defense, then other bacterial defense systems may hide the next generation of gene-editing tools.
The MIT-linked system, DefensePredictor, uses machine learning to scan bacterial genomes for proteins likely involved in anti-phage defense. It looks not only at a protein’s sequence, but also at its genomic neighborhood, because defense genes often cluster near one another. In 69 diverse E. coli strains, it predicted hundreds of previously unknown immune systems; researchers tested many of them and validated defense activity in dozens. Across 1,000 bacterial genomes, the model found nearly 3,000 protein clusters unlike known defense systems.
So the real breakthrough is not one new CRISPR. It is a search engine for nature’s hidden molecular machinery.
Why this matters
Older discovery methods were like walking through a forest with a candle. Scientists searched for known motifs or genes located in “defense islands.” DefensePredictor is more like using satellite imaging: it can detect patterns humans may not recognize.
This matters because bacterial defenses often have three qualities that genetic engineers love:
- Specificity — they recognize particular DNA or RNA sequences.
- Programmability — the target can sometimes be changed by changing a guide RNA or recognition sequence.
- Molecular action — they cut, paste, silence, signal, or destroy nucleic acids.
That is the recipe for a biotechnology tool.
The wider toolbox: how modern gene-engineering tools work
| Tool | How it works | What it can do | Example / evidence |
|---|---|---|---|
| CRISPR-Cas9 / Cas12 | A guide RNA brings a Cas enzyme to a matching DNA sequence; the enzyme cuts DNA; cellular repair then disrupts, removes, or replaces sequence. | Knock out genes, repair mutations, engineer immune or stem cells. | Casgevy, the first FDA-approved CRISPR/Cas9 therapy, edits blood stem cells to increase fetal hemoglobin in sickle-cell disease; FDA reported 29 of 31 evaluable patients were free from severe vaso-occlusive crises for at least 12 months. |
| Base editors | A disabled or nicking Cas protein carries a chemical editor, such as a deaminase, to one DNA letter and converts it without making a full double-strand break. | Single-letter correction, such as A→G or C→T changes. | VERVE-101 uses an adenine base editor targeting PCSK9 in liver cells; early human data showed dose-dependent PCSK9 and LDL-C reductions, though safety monitoring remains essential. |
| Prime editors | A nicking Cas9 is fused to reverse transcriptase. A pegRNA both finds the site and carries the new genetic “text” to be written. | More flexible “search-and-replace”: substitutions, small insertions, small deletions. | The original Nature paper reported more than 175 edits in human cells, including corrections relevant to sickle-cell disease and Tay–Sachs disease, without requiring double-strand breaks or donor DNA templates. |
| CRISPRi / CRISPRa | A dead Cas9, unable to cut DNA, is fused to repressors or activators and guided to promoters or enhancers. | Turn genes down or up without changing DNA sequence. | These tools are widely used in functional genomics screens to discover which genes drive disease states. CRISPRi represses transcription; CRISPRa activates transcription. |
| Epigenome editors | DNA-binding proteins such as dCas9, zinc fingers, or TALEs carry enzymes that add or remove epigenetic marks, such as DNA methylation. | Durable gene silencing without changing the DNA letters. | A Nature Medicine study designed an epigenetic editor targeting human PCSK9, inducing DNA methylation and near-complete PCSK9 silencing in transgenic mice after one lipid-nanoparticle treatment. |
| Cas13 / SHERLOCK-type tools | Cas13 targets RNA. When it finds its target, it can activate collateral RNA cleavage, which can be turned into a diagnostic signal. | RNA detection, viral diagnostics, mutation detection, RNA knockdown. | SHERLOCK uses Cas13 to detect viral genomes, cancer mutations, or antibiotic-resistance genes; improved versions can detect multiple targets and produce paper-strip readouts. |
| OMEGA systems: TnpB, IscB, IsrB | Small RNA-guided enzymes associated with mobile genetic elements. They may be evolutionary relatives or ancestors of CRISPR systems. | Potential compact genome editors, easier to deliver than large Cas9. | MIT/Broad researchers described OMEGA systems as RNA-guided DNA-modifying enzymes; some are about 30% the size of Cas9 and have been engineered to work in human cells. |
| Fanzor | A eukaryotic RNA-guided DNA-cutting enzyme related to OMEGA/TnpB systems. | Potential compact editor from complex life, not only bacteria. | Nature reported Fanzor as a eukaryotic programmable RNA-guided endonuclease that can be reprogrammed for human genome engineering. |
| Bridge recombinases | A “bridge RNA” has two programmable loops: one binds the target DNA, the other binds donor DNA. A recombinase performs insertion, excision, or inversion. | Large DNA rearrangements without relying on classic CRISPR cutting and repair. | A 2024 Nature study showed IS110 bridge RNAs can be independently reprogrammed to direct DNA insertion, excision, and inversion. |
| CRISPR-associated transposases / CASTs | CRISPR-like targeting guides transposase machinery to insert DNA cargo into a chosen genomic site. | Larger gene insertion, potentially useful for diseases caused by missing or broken genes. | A 2025 Science study reported evolved CAST variants with more than 200-fold improved activity and about 10–30% kilobase-size integration efficiencies in human cells at tested sites. |
| Retrons | Bacterial elements that make RNA-DNA hybrid molecules using reverse transcriptase; in nature they can act as phage-defense sensors. | Genome engineering, barcoding, recombineering, cellular recording. | Retrons were shown to function as anti-phage defense systems and generate RNA-DNA hybrids; this biology has inspired tools for genome writing and molecular recording. |
| CBASS, Thoeris, Zorya, SPARDA and related defenses | These systems often sense infection and activate signaling, toxicity, DNA/RNA destruction, membrane disruption, or abortive infection. | Future diagnostics, kill switches, biosensors, synthetic-biology control circuits. | Reviews describe CBASS, Pycsar, Thoeris, and type III CRISPR as nucleotide-signaling immune systems; Thoeris uses TIR-domain signaling and NAD-related mechanisms, linking bacterial immunity to broader innate immune biology. |
A simple metaphor
Think of genome engineering as editing a book:
- Cas9/Cas12 are scissors: cut a page and let the cell repair it.
- Base editors are pencils: change one letter.
- Prime editors are word processors: rewrite a short phrase.
- CRISPRi/a are dimmer switches: turn a chapter down or up.
- Epigenetic editors are bookmarks and locks: silence a chapter without changing the text.
- Bridge recombinases and CASTs are cut-and-paste systems: move or insert larger paragraphs.
- Cas13/SHERLOCK is a detector: it finds a sentence in RNA and raises a signal.
- DefensePredictor is the librarian: it searches nature’s hidden shelves for new tools.
What MIT’s discovery could lead to
The newly found proteins may become:
1. Smaller editors Cas9 is large, which makes delivery difficult. OMEGA and Fanzor-like systems may be smaller and easier to package into viral vectors or nanoparticles.
2. More precise nucleases Some bacterial proteins may recognize DNA or RNA in ways that reduce off-target effects.
3. New diagnostics Cas13 and Cas12 became diagnostic tools because of collateral cleavage. Other bacterial defense enzymes may produce signals that can be converted into cheap tests.
4. Programmable kill switches Toxin-antitoxin or abortive-infection systems could be adapted to prevent engineered microbes from spreading outside intended environments.
5. Large DNA writing tools Bridge recombinases and CASTs point toward a future where scientists can insert whole gene-sized payloads more precisely than with classic CRISPR repair.
6. Synthetic biology circuits CBASS, Thoeris, Zorya, and related systems behave like biological alarm systems. In engineered cells, similar logic could be used to make cells sense disease signals and respond.
The evidence is strong, but the promise is still early
There is solid evidence that bacterial defense systems can become revolutionary tools: CRISPR itself already became an FDA-approved therapy through Casgevy. There is also strong laboratory evidence that prime editing, base editing, bridge recombination, Fanzor, OMEGA systems, and CASTs can perform useful genetic operations.
But most of the DefensePredictor hits are still candidates, not ready-made therapies. Each must pass several hard tests: Can it be programmed? Does it work in human cells? Can it be delivered safely? Does it avoid off-target activity? Does the immune system tolerate it? Does it work in animals and then humans?
The beauty of the MIT work is that it changes the speed of discovery. Instead of waiting months to find a few candidates, AI can now reveal hundreds or thousands of possibilities rapidly. The future of gene editing may therefore be less like inventing tools from scratch and more like mining evolution’s ancient workshop—where bacteria, under viral attack for billions of years, have already built molecular machines we are only beginning to understand.