How CRISPR cuts and edits DNA

Bacteria have been fighting viruses for billions of years, and somewhere in that long war they invented a filing system. When a virus injected its DNA into a bacterium and the bacterium survived, it would clip out a small fragment of the invader's genetic code and paste it into a special region of its own genome called a CRISPR array — Clustered Regularly Interspaced Short Palindromic Repeats. The next time that virus came around, the bacterium could transcribe the stored fragment into a short piece of RNA, hand it to a protein called Cas9, and send the pair out hunting. The RNA was a wanted poster. Cas9 was the enforcer.

This system, repurposed in the laboratory, is what we now call CRISPR gene editing. The basic mechanism is strikingly simple for a tool this powerful. A researcher designs a guide RNA — a short sequence, usually about 20 nucleotides long — that matches the stretch of DNA they want to edit. The guide RNA is loaded into Cas9, and the complex is delivered into a cell. Inside the nucleus, Cas9 begins scanning the genome, briefly unzipping short stretches of the double helix to check whether the exposed bases pair with the guide. Most of the time, they do not, and Cas9 moves on. When it finds a match, it clamps down and makes a clean cut: a double-strand break, severing both strands of the DNA at almost exactly the same point.

There is one extra requirement worth naming. Cas9 will only cut if the matching site sits next to a short signature sequence called a PAM — a protospacer adjacent motif. The PAM is the bacterium's way of telling self from non-self: its own stored fragments in the CRISPR array do not have a PAM next to them, so Cas9 leaves them alone. In the lab, the PAM constrains where in a genome an edit can be aimed, but variants of Cas9 with different PAM preferences have widened the available targets considerably.

Cutting the DNA is only half the story. The cell does not tolerate a double-strand break; it rushes to repair the wound, and the repair pathway is what actually produces the edit. The faster, sloppier pathway is called non-homologous end joining. It glues the broken ends back together, often inserting or deleting a few bases in the process. Those tiny insertions and deletions, called indels, frequently shift the reading frame of a gene and knock it out. If the goal is simply to disable a gene — to ask what happens when this protein is missing — non-homologous end joining is usually enough.

The more precise pathway is homology-directed repair. If the researcher supplies a template DNA molecule whose ends match the sequences flanking the cut, the cell can use that template to rebuild the broken region. Whatever sequence sits in the middle of the template — a corrected version of a disease-causing mutation, a new tag, a different codon — gets written into the genome. Homology-directed repair is far less efficient than end joining, and it works mainly in dividing cells, which is one reason editing some tissues remains hard.

What makes CRISPR feel like a revolution, compared to earlier gene-editing tools, is the division of labor. Older systems required engineering a new protein for each new target — a slow, expensive craft. With CRISPR, the protein stays the same; only the 20-nucleotide guide changes. Targeting becomes a matter of typing a sequence rather than building a machine. The bacterium's filing system, in other words, turned out to be programmable, and the program is something you can order by mail.