Standard CRISPR-Cas9 cuts both strands of DNA and lets the cell patch the break, base editing swaps one DNA letter for another without a full cut, and prime editing rewrites short stretches of sequence using a built-in template. That is the whole family tree in one sentence. The differences matter because each tool fails in its own way, and those failure modes decide which one a research team, or a drug developer, reaches for.

All three share the same guidance trick. A short RNA leads a Cas protein to a matching address in the genome. What happens at that address is where they part ways.

The short version, side by side

Feature Cas9 nuclease Base editor Prime editor
What the protein does Cuts both DNA strands Cuts at most one strand; chemically converts one base Nicks one strand; copies new sequence from an RNA template
Typical outcome Small insertions or deletions (indels) that disable a gene A single-letter change inside a short window A programmed substitution, insertion or deletion
Needs a donor DNA template? Only for precise repair, and that route is inefficient No No (the template rides on the guide RNA)
Best suited to Switching a gene off Correcting or creating specific point mutations Edits base editors cannot reach, and small insertions or deletions
Main caveat Double-strand breaks carry risks Neighboring letters can change too Larger machinery, harder to deliver, edit efficiency varies by site

Read the table as tendencies, not rules. Engineers keep widening each tool’s reach, so a limit that holds this year may loosen the next. Still, the underlying logic stays put: the fewer strands you break and the more of the edit you spell out in advance, the more predictable the result, and the more machinery you must carry into the cell to get there.

Cas9 nuclease: the cut-and-let-the-cell-repair approach

The original tool is a molecular scissors. Cas9 is steered by a guide RNA, finds its target next to a short recognition motif, and cuts both strands. The cell then repairs the break. Most of the time it glues the ends back together in a sloppy way that adds or removes a few letters, and if that happens inside a gene the protein usually stops working.

That is a feature when the goal is to knock something out. It is a nuisance when the goal is precision. The cell can be coaxed into a cleaner repair by supplying a donor DNA template, but that homology-directed route is less efficient than the sloppy one, and the break itself is a stress on the cell.

The first gene-editing medicine approved anywhere works this way. Casgevy, from Vertex and CRISPR Therapeutics, is made by editing a patient’s own blood stem cells outside the body. Vertex’s first-quarter 2026 report lists it as approved for people 12 and older and reported $43 million in quarterly Casgevy revenue, which gives a sense of how early the commercial story still is (Vertex, Q1 2026 results).

Base editing: changing one letter without a full cut

David Liu’s group at the Broad Institute described the first base editors in 2016. They fused a Cas9 that no longer cuts both strands to an enzyme that chemically converts one base into another. The original design turned a C into what the cell reads as a T, with the matching change on the opposite strand. The report described editing in roughly 15 to 75 percent of DNA in cultured cell lines, with indels typically at or below 1 percent (Broad Institute, Komor et al. 2016 record; PubMed entry).

Later editors added the reverse chemistry, turning A into G. Together the two families cover what are called transition mutations, which is a large share of known disease-linked point mutations but not all of them. Their editing window spans only a handful of letters, and if the same letter appears twice inside the window, both can change. Those bystander edits are the practical headache of base editing.

Because no break across both strands is needed, the approach avoids most of the indel mess. That is one reason several in vivo programs, including the cholesterol program Lilly acquired with Verve, use base editors.

Prime editing: a search-and-replace function

Prime editing, reported by Anzalone and colleagues in 2019, takes a different route. A Cas9 that nicks only one strand is joined to a reverse transcriptase, an enzyme that writes DNA from an RNA template. A longer guide, the prime editing guide RNA or pegRNA, does two jobs: it finds the address and it carries the new sequence to be copied in. The paper described no double-strand breaks and no donor DNA (Anzalone et al., Nature 2019, via PubMed Central).

The authors reported more than 175 edits in human cells, covering all 12 possible point-mutation swaps plus small insertions and deletions. They estimated that the approach could in principle address up to 89 percent of known disease-associated genetic variants. Those figures describe what is possible in a dish, not what any single clinical product can do. They also reported far lower off-target editing than Cas9 nuclease at known problem sites.

The cost is complexity. A prime editor is a large protein plus a long guide, and that is harder to package for delivery. Efficiency also swings from one target to the next, which is why teams spend effort screening guide designs.

Which one has human data?

Each tool now has some clinical evidence, though at very different depths.

  • Cas9 nuclease: an approved ex vivo therapy (Casgevy) and in vivo programs such as Intellia’s lonvo-z, which posted phase 3 results in 2026, based on company announcements and a journal publication.
  • Base editing: Lilly reported interim phase 1b data for VERVE-102 in May 2026, and a personalized base editor treated an infant with CPS1 deficiency in 2025.
  • Prime editing: Prime Medicine reported results from two patients with chronic granulomatous disease in December 2025, with DHR-positive neutrophils at 69 percent and 83 percent by day 30. That is a company release alongside a New England Journal of Medicine paper, and two patients is a very small sample (Prime Medicine press release, December 7, 2025).

Notice that the prime editing case was an ex vivo treatment, where cells are edited outside the body and returned. Delivering a prime editor to organs inside the body is the next test, and preclinical liver work has used a two-part system, a lipid nanoparticle carrying editor mRNA plus a viral vector carrying guides (CHOP and Penn, March 31, 2026).

What the edit efficiency numbers do and do not tell you

Editing percentages are easy to misread. A figure like 30 to 40 percent correction, which the CHOP and Penn team reported for a prime editing approach in liver cell and animal preclinical work, sounds modest. The same team considers roughly 10 percent correction necessary for clinical benefit in that disease, so for them it clears the bar by a wide margin. The threshold depends on the disease, on how many cells must be corrected, and on whether corrected cells gain a growth advantage.

Cell-culture numbers also run higher than what you can expect in a living animal, because culture dishes are easy to reach. The honest comparison is always the one measured in the setting that matches the intended therapy.

Myths worth retiring

Myth: base and prime editors never cause unintended changes. They avoid double-strand breaks, which removes one class of damage, but they can still edit the wrong site or the wrong neighboring letter. How researchers look for such errors is a topic of its own.

Myth: newer means better for every job. If the goal is to disable a gene, a simple Cas9 cut is often the more direct tool. Precision tools earn their keep when the exact letter matters.

Myth: one edit equals a cure. A correct edit in the right cell type is necessary, not sufficient. Enough cells must be reached, and the edit must change the disease in a way patients feel.

How to choose, in plain terms

  1. Start with the change you want. A knockout points to Cas9 nuclease.
  2. If it is a single-letter swap of the right type, check whether a base editor’s window would also hit neighboring letters.
  3. If the change is a transversion, a small insertion or a small deletion, a prime editor is the likelier fit.
  4. Then ask how the editor will physically reach the cells. Delivery often decides the winner, and that is where the packaging limits above start to bite.

This is general scientific background, not medical advice, and trial results discussed here are early and company-reported unless a journal paper is cited.

Sources