And an article in The Atlantic that some may find more approachable than the original Nature Communications publication:
Cas9 (and similar proteins) have 2 basic functions:
1) to bind to specific DNA (by means of complementing a specifically-bound RNA)
2) to cut the DNA that is has bound.
Only one of these functions is actually desirable or unique to Cas9 - that of binding to a sequence-specific part of DNA. The 'cutting' function is how the CRISPR system naturally destroys foreign DNA in Cas9's normal environment, but is generally not desirable in these synthetic biology or therapeutic applications. For genetic engineering purposes we really have better tools than Cas9 to cut, modify, or otherwise affect DNA - but Cas9 is clearly a fantastic tool to localize to particular sequences. That this paper so clearly demonstrates the terrifying damage that cleavage can do to genomic DNA only reinforces that Cas9 is a useful scaffold upon which to engineer further tools, but in its wild-type form is kind of scary.
Recent papers from the Liu lab at Harvard  (among many others) show how you can edit DNA using a modified Cas9 that has been modified to not have that second cutting function, while retaining the ability to bind to a specific DNA sequence. The synthetic "BaseEditor" protein has been further modified to have more complicated DNA modification capabilities, rather than whole-sale slicing of DNA . Though this designed protein with Cas9 at its core is complicated, it should not cleave and let dangle genomic DNA as we see it here visualized. I'd be curious to see how the BaseEditors behave under a similar visualization.
 Kim et al. 2017: https://www.nature.com/articles/nbt.3803
 BaseEditor v4: https://serotiny.bio/notes/proteins/sabe4gam/
"Possibility" is too strong a term. Afaict, the best you can say about repair is that it is a "less than common occurrence," and even that may be too much.
>"Recent papers from the Liu lab at Harvard  (among many others) show how you can edit DNA using a modified Cas9 that has been modified to not have that second cutting function, while retaining the ability to bind to a specific DNA sequence."
Sorry but I could not find any comment/data about toxicity in this paper. This is basic info everyone wants to know: What percent of cells exposed to this stuff die? Maybe I missed it?
That experiment has been done and is actually written up here as a protocol to more effectively using Cas9. The paper clearly refutes the above hypothesis and demonstrates that modifications by Cas9 are generally proportional to the amount of expressed Cas9 up to the point of toxicity . In it they tag Cas9 with a fluorophore, express the construct in cells and sort for those cells that express Cas9 (separating them from cells that do not), and binning the cells based on the level of Cas9 expression. Given your hypothesis, those fluorescent cells should die and not have the desired mutations, while the non-fluorescent cells will be enriched for particular mutations not found in the fluorescent cells. Some toxicity is shown at very high levels of Cas9 expression, but the on-target edits occur in cells expressing Cas9, and those cells live for a number of passages. See Figure 3 in the paper below:
> Cell population 1 displays GFP fluorescence intensities similar to those of nontransfected cells (autofluorescence level), whereas populations 2–6 display specific GFP levels ranging from low to high (varying ~500-fold). (e) 20,000 cells were isolated from each of populations 1–6 in (d) and cultured for 6 d, after which IDAA was used to determine indels at the KRAS locus in an aliquot of the cells. Note that editing levels increase with increasing GFP fluorescence level at the time of FACS and that almost complete editing was obtained at submaximal fluorescence levels... [T]he total number of live cells in each population analyzed in (e) was determined by trypan blue exclusion and expressed as % of population 1.
Given the above result, a significant amount of research is being done right now to preserve the useful properties of Cas9 - being able to home in on a particular genetic sequence - while engineering the protein to minimize and even prevent those deleterious cuts do DNA. And that research is finding steady success.
The hypothesis is that it is more likely to kill those cells than the others, and does so at a high enough rate to select for cells missing the target site. Other things such as NHEJ/HDR happen too.
Anyway, the technique in the paper you linked is not supposed to be causing double strand breaks, which for a long time before CRISPR/Cas-9 were thought to result in Death >> NHEJ >> HDR. This is a totally different scenario, but it is quite odd they do not mention toxicity at all since that is a major concern of these technologies.
I'll check the Lonowski et al paper and get back to you.
> "Given your hypothesis, those fluorescent cells should die and not have the desired mutations, while the non-fluorescent cells will be enriched for particular mutations not found in the fluorescent cells"
Perhaps there is a typo here but this is exactly the opposite of what my hypothesis predicts. The surviving cells should contain cas-9 and the mutation. Cas-9 damages the DNA of some cells less because the the mutation was already there. The cells with no cas-9 are noise because they were never treated.
If the mutations were already there, then control in figure 7 in Lonowski et al. would not contain 0% "on-target" mutations. The presence of Cas9 is clearly causing a genetic change that is not already present in the population, and those cells that are changed do not all die. It is clearly not the case that the cells with the desired mutations completely overlapped with the cells transfected.
So they do see mutations in the control cells but decided to call it some kind of artefact. BTW, I also see a less shoulder in the left (on-target) site of the same figure.
The point is that these mutants pre-exist in low frequencies of 1/100 to 1/100,000 cells (depends on cell type, environment, etc), almost regardless of the target site. If you look at enough cells you are guaranteed to find one with an indel at any site. In those cases, just as cas-9 doesn't just keep re-editing the dna after the target sequence has been cut, the cells are relatively "immune" to it.
So for figure 7 they started with 50k cells, the questions are:
1) How many of those cells already contained a mutation at the target site?
2) What was the growth curve of the cells after treatment?
3) What is the division rate of these cells under these conditions?
4) What is the death rate of these cells under these conditions?
5) How many cells were there in the end for final analysis (sequencing, whatever)?
EDIT: Found one here: .