Lost in Latent Space

Chapter: From PAM to Repair — The Mechanism of CRISPR-Cas9 Editing

1. Overview

The phrase “CRISPR edits DNA” hides three distinct events: target recognition, double-strand cleavage, and repair. Only the first two are performed by the Cas9 ribonucleoprotein. The third — the one that determines whether you get a clean substitution, a random insertion, or no change at all — is performed by the host cell’s repair machinery, which the experimenter does not directly control. This chapter walks through each step in order, names the structures involved, and ends with the question every editing experiment ultimately turns on: which repair pathway will the cell choose?

2. The guide RNA and the PAM

Cas9 is an endonuclease. On its own it does nothing useful; it must be loaded with a single-guide RNA (sgRNA) that specifies the target. The sgRNA contains two functional regions:

  1. A scaffold that folds into a structure Cas9 recognizes.
  2. A spacer — typically 20 nucleotides — whose sequence is complementary to the desired DNA target (the protospacer).

A common point of confusion: the spacer does not, by itself, suffice for binding. Cas9 will only cut DNA where the protospacer is immediately adjacent (on the 3’ side) to a short motif called the PAM (protospacer adjacent motif). For Streptococcus pyogenes Cas9 — the default workhorse — the PAM is 5'-NGG-3'. The PAM is read on the non-target strand and is recognized directly by Cas9 protein contacts, not by the guide RNA.

The PAM exists because evolution had to solve a problem: CRISPR originated as a bacterial immune system. Bacteria store fragments of viral DNA in their own genome (in CRISPR arrays) and transcribe them into guides. The PAM is present in the viral DNA but, by design, absent from the bacterium’s own CRISPR array. This is how Cas9 avoids cleaving its own host. For genome editing, the PAM is mostly a constraint on which sites you can target: roughly one NGG every 8 bp in a random sequence.

3. Target search and R-loop formation

Cas9 does not scan DNA linearly. It diffuses through the nucleus and tests sites by first asking, “Is there a PAM here?” PAM recognition takes microseconds; the protein interrogates many sites per second.

Only after a PAM is found does Cas9 begin to unwind the adjacent DNA and test for complementarity with the spacer. Pairing initiates at the seed region — the ~10–12 nucleotides of the spacer closest to the PAM — and propagates outward. If the seed mismatches, Cas9 dissociates almost immediately. If the seed pairs but distal nucleotides mismatch, Cas9 may still bind but will often fail to cleave. This is the structural origin of CRISPR’s specificity: errors near the PAM are catastrophic, errors far from the PAM are forgivable. Practitioners exploit this when scoring off-target risk.

Successful pairing creates an R-loop: the spacer RNA hybridized to the target DNA strand, with the non-target strand displaced as single-stranded DNA. R-loop formation triggers a conformational change in Cas9 that positions two distinct nuclease domains for cutting.

4. The double-strand break

Cas9 cuts each strand with a different active site:

Both cuts occur 3 bp upstream of the PAM, producing a blunt-ended double-strand break (DSB). After cleavage, Cas9 remains bound to the broken ends for tens of seconds to minutes — a kinetic detail that matters because it shapes which repair factors arrive first.

A useful exercise: given the sequence below, identify the cut site.

5'-...GACATTGGCCAATCATCGAGCAA-NGG-...-3'   (non-target / PAM-bearing strand)
3'-...CTGTAACCGGTTAGTAGCTCGTT-NCC-...-5'   (target strand)

If the PAM is the NGG on the top strand, the cut is between the 17th and 18th nucleotide of the spacer counted from the PAM-proximal end. Both strands are cut at the same position. The result is a blunt DSB.

5. Repair pathway choice

A blunt DSB is a serious lesion. The cell must repair it or die. There are two dominant pathways:

5.1 Non-homologous end joining (NHEJ)

NHEJ is fast, active throughout the cell cycle, and does not require a template. Ku70/Ku80 binds the broken ends; DNA-PKcs recruits processing enzymes; ligase IV seals the ends. The drawback is that NHEJ is error-prone: nucleotides are frequently inserted or deleted at the junction (indels). At a typical Cas9 cut site, NHEJ produces a distribution of small indels — often 1 bp insertions and 1–10 bp deletions, with the exact spectrum biased by local sequence (microhomologies in particular).

If your goal is to knock out a gene by frame-shifting its coding sequence, NHEJ is your friend: roughly two-thirds of random indels will produce a frameshift, and the gene’s product will likely be nonfunctional.

5.2 Homology-directed repair (HDR)

HDR uses a homologous DNA template — natively, the sister chromatid; experimentally, a donor DNA you co-deliver — to copy a sequence across the break. HDR is precise: you can install a specific point mutation, insert a tag, or replace an exon, as long as your donor’s homology arms match the flanking genome.

HDR has two large drawbacks. First, it is restricted to S and G2 phases of the cell cycle, because it relies on machinery shared with replication. Second, it is inefficient relative to NHEJ — typically <10% of edited alleles in primary cells, often <1%. This is why HDR knock-ins are still considered hard, even a decade into the CRISPR era, and why entire sub-fields exist to bias the pathway choice (small-molecule NHEJ inhibitors, cell-cycle synchronization, single-stranded donor templates, base editors that bypass DSBs entirely).

6. Worked example: knock-out vs knock-in

Suppose you want to disrupt the gene EMX1 in a human cell line.

The asymmetry between these workflows is not a quirk; it is downstream of the pathway-choice problem. Anything you do to push the cell toward HDR trades efficiency, viability, or cell-cycle range.

7. What’s actually being “edited”

A useful frame: Cas9 does not edit DNA. Cas9 breaks DNA. The cell edits DNA. Everything called “CRISPR engineering” downstream of base/prime editors still bottlenecks on this fact. Once you have grasped target search, R-loop formation, cleavage, and the dichotomy of repair pathways, you have the conceptual core. The rest of the field — Cas12, Cas13, base editors, prime editors, epigenome editors, CRISPRi/a — is permutations on which step (recognition, cleavage, repair) you change.

Test your understanding

  1. Why does CRISPR-Cas9 require a PAM, biologically speaking?
  2. A guide has perfect complementarity to the protospacer except for two mismatches at the PAM-distal end. Will Cas9 cleave? Why or why not?
  3. You introduce a Cas9 + sgRNA + ssDNA donor into a non-dividing neuron. You see no HDR products. Give the most likely mechanistic reason.
  4. Distinguish, in one sentence each, the roles of the HNH domain and the RuvC domain.

What’s next

The natural follow-up is base editing — a Cas9 variant that nicks rather than cleaves and chemically converts one base into another without ever inducing a DSB. It sidesteps the repair-pathway lottery entirely, which is why it has become the preferred tool for installing single point mutations. After that, consider prime editing, which uses a fused reverse transcriptase to install arbitrary small edits — a much richer mechanism but with its own efficiency story.