# After the Cut: How DNA Repair Decides Between a Knockout and a Precise Edit

Why end joining and HDR give opposite results from the same break, and where base and prime editing fit in

> DNA repair · base editing · prime editing · NHEJ and HDR · About 10 min · Oct 6

## Key points

1. CRISPR’s target search and cut do not determine the edit — the cell’s repair of the double-strand break does, and the same cut can end in either a destroyed gene or a precise correction.
2. End joining stitches broken ends without a template, works throughout the cell cycle, and produces small random indels; an indel whose length is not a multiple of three shifts the reading frame and usually yields a knockout.
3. Because a perfectly rejoined site can be cut again, repeated cutting biases outcomes toward lasting indels — but an indel is not automatically a null allele, and both gene copies must be checked.
4. HDR rebuilds a break accurately by copying from a template, and in the lab the donor template carries the sequence you want; it is limited to S and G2, weak in non-dividing cells, and competes with end joining.
5. HDR efficiency depends on cell type, genomic location, and nuclease rather than being a fixed property of CRISPR, and indel frequency does not reliably predict HDR frequency; blocking end joining (e.g., ligase IV or DNA-PK inhibitors) can raise precise insertion many-fold.
6. Base editors avoid breaks entirely: an impaired Cas9 unzips a bubble, a fused deaminase converts C to U (read as T) or A to inosine (read as G), and a nick biases repair, confining editing to roughly positions 4–8 and risking bystander edits; only transitions are possible.
7. Prime editing nicks the PAM-containing strand, and the freed DNA end is extended by reverse transcriptase using a prime editing guide RNA as the template, writing the desired sequence directly with all twelve substitutions plus small insertions and deletions of at least dozens of bases.
8. The most precise method for a job depends on the outcome needed: knockout asks for end joining, letter correction asks for base editing or HDR, and small insertions, deletions, or transversions point to prime editing.
9. The first approved CRISPR therapy, Casgevy, is a deliberate knockout of the BCL11A erythroid enhancer repaired by end joining, while the newer risto-cel raises fetal hemoglobin with a base edit of the HBG1/HBG2 promoters — showing that “more precise” and “first to the clinic” are different questions.

---

In the previous article we followed Cas9 to the moment of cutting: a 20-letter guide RNA pairs with one DNA strand, the PAM next to it is read by the protein, and two nuclease domains snip both strands. That is where the protein’s job ends.

Everything you actually wanted — a broken gene, a corrected letter, an inserted tag — is done afterwards, by the cell’s own repair machinery, using its own tools. And that machinery offers more than one option.

This is the single most important idea in this article: CRISPR does not edit DNA. It creates a problem, and the way the cell solves that problem is what becomes the edit. The same cut, in the same gene, in the same cell type, can produce a destroyed gene or a letter-perfect correction, depending only on which repair route the cell takes.

## A double-strand break is an emergency

When both strands of the helix are broken, the cell has lost the information it needs to reconstruct the missing piece — the opposite strand can no longer serve as a backup copy. So a double-strand break (DSB) is treated as an emergency, and cells have evolved two broad ways to handle it.

**End joining** takes the two broken ends and simply glues them back together. It needs no template, it works at any point in the cell cycle, and it is the dominant response in most human cells. It is also sloppy: the ends often get chewed back or filled in by a few bases before being ligated, so the repaired sequence is usually not identical to the original.

**Homology-directed repair (HDR)** is the careful route. Instead of gluing, the cell searches for a matching sequence to copy from and uses it as a template to rebuild the broken region accurately. The catch is where that template comes from: normally it is the sister chromatid, which only exists after the DNA has been replicated. So HDR is essentially limited to the S and G2 phases of the cell cycle, while end joining is available throughout.

Both systems are active at once and compete for the same break. Which one wins is the whole game.

## Route one: glue the ends, lose the gene

If you cut inside a gene and let end joining take over, the result is a small insertion or deletion — collectively called an **indel**, typically a handful of bases, sometimes more. The repair is random-ish in length, and that randomness is the point.

A gene’s coding sequence is read in groups of three bases, one group per amino acid. Add or remove a number of bases that is not a multiple of three, and every group downstream is shifted by one or two letters: the reading frame is broken. From that point on the sequence spells a different set of amino acids, almost always running into a premature “stop” signal within a short distance. The cell’s quality control usually destroys such a truncated protein, and the gene is effectively off. That is a **knockout**.

Two wrinkles are worth knowing, because they explain why knockout experiments need checking rather than assuming:

- End joining sometimes restores the original sequence exactly. That restored site is now a perfect target again, so if Cas9 is still around it will cut a second time, and a third. Over time, alleles that survive tend to be the ones carrying indels. The cut keeps re-opening the door until it produces a lasting mistake.
- Not every indel is a knockout. A deletion of three or six bases leaves the frame intact and removes or adds a few amino acids; the protein may still work. And most cells are diploid, so unless both copies of the gene are disrupted, the cell may still make protein from the untouched allele. “I got an indel” and “I got a null allele” are different statements, and only sequencing plus a protein-level check can tell you which one you have.

End joining is not one single pathway, either. Besides the canonical machinery that ligates the ends, a backup route called microhomology-mediated end joining uses very short stretches of accidental matching sequence near the break, and it tends to produce larger deletions. Practically, this is why the exact indels you get at a site are somewhat predictable but never guaranteed.

## Route two: copy from a template, get a precise edit

If instead you want to *write* something at the cut — replace a wrong letter, insert a tag, correct a mutation — you need HDR, and HDR needs a template. In the lab you supply one: a **donor template**, a piece of DNA whose middle carries the sequence you want and whose ends match the DNA on either side of the cut. The cell resects the broken ends, lets the donor’s matching ends find the site, and copies your sequence into the gap.

Now the outcome is entirely up to the letters you wrote into the donor. This is how the sickle-cell mutation in *HBB*, a single A-to-T swap, can in principle be reverted to the healthy sequence, and how a whole fluorescent protein or a corrected gene copy can be dropped into a chosen location.

The problem is efficiency and cleanliness. HDR is competing with end joining for the same break, and in most therapeutically relevant human cells end joining wins. Measured HDR rates run from a few percent to a few tens of percent depending on the cell type, the exact location in the genome, and which nuclease you use — the ratio is not a fixed property of CRISPR, it is a property of your particular experiment [1][2]. Worse, the two outcomes do not even track each other: editing activity measured as indels does not reliably predict how much HDR you will get at the same site [2].

Non-dividing cells are the hard boundary. Neurons, muscle, and most mature liver cells are not cycling, so they have no sister chromatid to copy from and no meaningful HDR. If the target tissue is post-mitotic, precise editing by cutting plus a donor is largely off the table.

People do try to shift the balance. Blocking a key end-joining enzyme makes more breaks available to HDR; in one study a ligase IV inhibitor raised precise insertion rates up to about 19-fold in a human cell line, and a newer DNA-PK inhibitor is used as an HDR booster in primary cells [3]. Others restrict Cas9 activity to S/G2 so it only cuts when HDR is possible. All of these are ways of nudging a ratio that the cell, not you, sets by default.

## Two newer tools that skip the break entirely

The awkwardness above explains the design of the next generation of tools. If the two possible repairs of a break are “random small mistake” and “accurate but rare,” the obvious move is to stop making breaks.

**Base editing** does chemistry instead of cutting. Take the Cas9 targeting machinery we already know, disable its ability to cut both strands, and fuse it to an enzyme that modifies bases. When Cas9 binds, it unzips a short stretch of DNA and pairs the guide with one strand, leaving the other strand briefly single-stranded — the same bubble the previous article described. A **cytidine deaminase** attached to the protein acts on that exposed single strand, converting cytosine (C) into uracil (U). To a DNA polymerase, U looks like thymine (T). An accompanying nick on the opposite strand biases the cell’s mismatch repair to accept the change, and the pair C•G becomes T•A. The mirror-image tool, an **adenine base editor**, converts adenine (A) into inosine, which polymerases read as guanine (G), so A•T becomes G•C.

Two properties follow from this mechanism:

- The chemistry only happens on the single-stranded bubble, so the reaction is confined to a small **editing window**, roughly positions 4 to 8 of the 20-letter target site. Anything editable inside that window gets edited, which means if your target base has an editable neighbour, both change — a **bystander edit**. Designing a base-editing experiment is mostly about finding a window where the target sits alone.
- A base editor cannot delete, insert, or swap a purine for a pyrimidine. It performs only the four “transition” substitutions. That is a real limit, but transitions account for a large share of known disease-causing alleles [4]. In exchange, editing is remarkably pure — typically far more desired product than indel byproducts — and it works in cells that never divide, which is exactly where HDR fails [4].

**Prime editing** goes further while still avoiding a full break. Take a Cas9 that nicks only one strand — the strand carrying the PAM, which is the one the guide does not pair with — and fuse it to a reverse transcriptase, an enzyme that copies RNA into DNA. The guide RNA is extended into a **prime editing guide RNA (pegRNA)**, which carries not only the targeting sequence but a short primer-binding site and a template spelling out the change you want. After the nick, the freed DNA end anneals to the primer-binding site, and the reverse transcriptase copies the pegRNA’s template onto the DNA end — writing your sequence directly into the genome. The cell then resolves the resulting pair of flaps, removing the old sequence and leaving the new one. A second nick on the unedited strand (the “PE3” configuration) biases repair toward the edited strand and raises efficiency to roughly 20–50% in a commonly used cell line, with 1–10% indels [5].

Because the new information comes from a template you wrote into the guide, no donor DNA is needed, and there is no double-strand break to be randomly rejoined. Prime editing can install all twelve kinds of single-base substitution, plus small insertions and deletions — demonstrated for insertions of at least 44 bases and deletions of at least 80 [4][5]. In principle that covers up to about 89% of known disease-causing variants [5]. The trade-offs are size (the editor is a big protein, which complicates delivery) and efficiency, which is often lower than base editing and highly dependent on the site.

```mermaid
flowchart LR
  A["Cas9 cuts both strands"] --> B{"Which repair route?"}
  B --> C["End joining: no template, any cell cycle"]
  C --> D["Random small indels"]
  D --> E["Reading frame shifts - gene knocked out"]
  B --> F["HDR: donor template, S and G2 only"]
  F --> G["Your donor letters copied in - precise edit"]
  H["No break: base editing"] --> I["Deaminase on the unzipped bubble - C to T or A to G"]
  J["No break: prime editing"] --> K["Nickase plus reverse transcriptase - pegRNA writes new sequence"]
```

## The decision, and a revealing clinical example

Put together, the choice of tool is really a choice of repair outcome:

- **Destroy a gene?** Cut it and let end joining do the work. Fast, efficient, easy.
- **Rewrite a letter, using a donor?** Cut it, supply a donor, and fight for HDR — with the knowledge that it will be inefficient in any cell that is not dividing.
- **Rewrite a letter, cleanly?** Use a base editor if the change is a transition and a suitable window exists.
- **Insert or delete a few bases, or make a transversion?** Use a prime editor.

The first approved CRISPR medicine illustrates how counter-intuitive the mapping can be. Casgevy, a therapy for sickle cell disease and beta-thalassemia, works by cutting the erythroid-specific enhancer of *BCL11A* in a patient’s own blood stem cells. The breaks are then repaired by end joining, producing indels that destroy a binding site for the transcription factor GATA1. With less BCL11A, the cells re-express fetal hemoglobin, which does not sickle. In other words, the first approved CRISPR therapy is not a precise edit at all — it is a deliberate, deliberate-looking knockout, and the “sloppy” pathway is exactly what the therapy wants [6][7]. Editing frequencies in the manufactured cells run from about 60% to 92%, and treated patients maintain roughly 70% or more edited alleles with fetal hemoglobin above 40% of their total hemoglobin [7].

A newer sickle-cell candidate, risto-cel, does something related with a base editor: instead of breaking the *BCL11A* enhancer, it changes letters in the *HBG1* and *HBG2* promoters so that BCL11A can no longer bind there, again raising fetal hemoglobin without a double-strand break. In an early trial, patients reached a mean of about 67% on-target edited alleles at six months, with fetal hemoglobin above 60% and sickle hemoglobin below 40% of the total [8]. The reported trial also reminds us how young this field is: serious adverse events were common, and one patient died of idiopathic pneumonia syndrome [8].

## What this does and does not settle

The repair pathway explains the outcome of editing, but it does not remove the other practical constraints. Reagents still have to reach the right cells, in enough of them, and an organ is not a petri dish — editing a fraction of cells in a tissue produces a mosaic result. Cutting also carries risks that are not about the targeted letters at all: double-strand breaks have been associated with large deletions, complex rearrangements, and activation of the p53 damage response, which is part of why DSB-free methods such as base and prime editing attract attention beyond their precision [4].

And none of this decides the ethical questions. Whether editing should be confined to a patient’s own body cells or extended to embryos and sperm, and who gets to decide, are questions about purpose and consent, not about which repair enzyme runs. The biology tells you what the tool does. It does not tell you what to do with it.

## Sources

1. [Current Strategies for Increasing Knock-In Efficiency in CRISPR-Based Approaches — HDR versus NHEJ, cell-cycle dependence, and ways to shift the balance](https://pmc.ncbi.nlm.nih.gov/articles/PMC10931195/)
2. [Systematic quantification of HDR and NHEJ — the ratio depends on locus, cell type, and nuclease, and the two frequencies do not correlate](https://www.nature.com/articles/srep23549)
3. [Inhibition of non-homologous end joining increases HDR — ligase IV inhibitor Scr7 raised precise insertion up to ~19-fold](https://pmc.ncbi.nlm.nih.gov/articles/PMC4618510/)
4. [Genome editing with CRISPR–Cas nucleases, base editors and prime editors — indel outcomes, base-editing windows, prime-editing scope, and DSB-associated risks](https://www.cell.com/molecular-therapy-family/molecular-therapy/fulltext/S1525-0016(21)00457-3)
5. [Search-and-replace genome editing without double-strand breaks or donor DNA — the original prime editing paper (pegRNA, PE2/PE3, ~89% of disease variants in principle)](https://www.nature.com/articles/s41586-019-1711-4)
6. [Regulatory assessment of Casgevy — BCL11A enhancer cut repaired by NHEJ indels, and the downstream fetal hemoglobin mechanism](https://www.mdpi.com/1467-3045/46/8/485)
7. [FDA package insert for Casgevy — editing frequencies, patient allele editing levels, and fetal hemoglobin percentages](https://www.fda.gov/media/174615/download)
8. [Base Editing of HBG1 and HBG2 Promoters for Sickle Cell Disease — phase 1–2 results for risto-cel, including editing levels and adverse events](https://www.nejm.org/doi/full/10.1056/NEJMoa2504835)

---

Original article: https://eulore.ai/articles/dna-repair-after-crispr-cut-4dc6b8e4

> **Eulore** · Learn a little. Understand a lot.
>
> Eulore is an AI learning tool that turns what you want to learn into a continuing series. Share a topic, and it gets to know your starting point before creating articles you can read in 5–10 minutes. Ask as you read, and shape what comes next.This article was created in the same way.
>
> Start your own series → https://eulore.ai