Borrowed from bacteria
CRISPR did not start as a tool. It is a defense system that many bacteria use against viruses. When a virus attacks, a bacterium can file away a short snippet of the virus's DNA in its own genome — a kind of mug-shot library. If that virus returns, the bacterium copies the snippet into RNA and hands it to a cutting protein, which uses the snippet to find and slice the matching viral DNA. Scientists realized this natural “search-and-cut” machine could be re-aimed at any sequence we choose.
The best-known version uses a protein called Cas9. Think of Cas9 as the scissors from guide 1, and the RNA snippet as the GPS. Together they form a tiny guided missile: the RNA reads the target, the protein makes the cut. Crucially, *we* get to write the RNA — so we get to choose the target.
The guide RNA is the address
The targeting part is the guide RNA. In modern editing it's usually fused into one molecule, the single guide RNA, that does both jobs: it holds onto Cas9 and it carries a roughly 20-letter stretch that spells out the target. That 20-letter stretch finds its match by ordinary base pairing — A with U, G with C — the same rule that holds the DNA double helix together. Where the guide's letters pair up with one strand of the DNA, Cas9 knows it has arrived.
This is the elegant trick. To re-aim the whole machine at a new gene, you don't redesign the protein at all — you just type a new 20-letter guide sequence. That ease is why CRISPR spread through laboratories so fast: changing the target is as simple as changing a search term.
The PAM: a tag the cut depends on
There is one more rule, and it's easy to overlook. Matching the guide RNA is not enough on its own. Right next to the target, the DNA must also carry a tiny landmark called the protospacer adjacent motif, or PAM. For the common Cas9, the PAM is just three letters, usually written “NGG” — meaning any letter, then two G's. No PAM next door, no cut, even if the guide matches perfectly.
Why does this matter? Two reasons. First, the PAM is how Cas9 avoids cutting the bacterium's *own* CRISPR library — the stored snippets have no PAM, so they are spared. Second, for us it is a constraint: you can only edit close to a spot where a PAM exists. That's usually fine — short PAMs are common — but it explains why not every single letter in the genome is reachable by the basic tool, and why researchers keep finding Cas proteins with different PAM rules to widen the reach.
Target DNA (top strand): 5'-...AGGTCATCGGACTTGCAATGCA TGG ...-3'
|-- 20-letter target --| |PAM|
Guide RNA (matches target): 3'- UCCAGUAGCCUGAACGUUACGU -5'
Cas9 checks two things, in order:
1. Is there a PAM (NGG) next to the site? -> TGG yes
2. Does the 20-letter guide base-pair the target? -> yes
Both true -> Cas9 cuts ~3 letters upstream of the PAM:
...AGGTCATCGGACTTGCAA | TGCA TGG...
^ double-strand break here