One neat way might be to REPAIR the original mutation that you introduced with the CRISPR/Cas9 - using homology-directed repair with a donor template in combination with the same CRISPR/Cas9 - to see if it recovers the wild-type phenotype. Off-target effects would seem very unlikely to be correctly repaired by this mechanism.

What was the exact modification? If it was simply a knock out, targetting a different sequence within the same gene and verifying that the phenotypic effects are still the same should rule out off-targetting as the cause for that phenotype, since a different guide RNA should have a completeley different set of possible off-targets. Right?

Of course, whether the reviewers would live with it or not is another matter...

Hello Manuel

In my point of view, off-target is a real issue only if you want to develop therapies for human health. For research,  this question could also be addressed,  but in a less extent for the following reasons: 1) you don't have time and money to spend to completely analyze the whole genome for off-target 2) you work with an "artificial" system (I mean in vitro), so, by nature, greatly imperfect 3) and thid system is permanently changing. I think that the big issue is the selection of the right clone after genome modification (be sure to not select a "non-relevant" clone).

Functional characterization is sufficient to work.

Sebastien

Since CRISPRY/Cas9 system is developed for specific gene 'targeting', if it gets a lot of 'off-targets', then basically this system cannot be viewed as a powerful tool for targeting purpose. So, off-targets problem need to be addressed. Serious off-targeting can cause disturbance for the genome or even genome fracturation. I hope one day the price of deep sequencing can greatly reduced.

Hi Manuel,

It is possible that the reviewers worried that some unsuspected off-target mutations could be involved in some phenotypes observed using your cell lines. This could especially be true when those phenotypes are weak and/or could be affected by mutations in numerous targets (e.g. mutations affecting fitness or a stress pathway).

An alternative to deep sequencing to rule-out off-target mutations would be to independently reconstruct the cell lines (and, as mentioned by Alejandro, targeting a different sequence within the same gene) and show that their phenotype is consistent with that of the first cell lines constructed. However, this approach is time-consuming and, depending on the number of cell lines, the complexity of the mutations and the type of phenotypes observed, can become as (or more) expensive than deep sequencing.

However, if the cell lines have not been constructed for a specific study but for a more general use , off-target mutations might become a problem in future studies using those cell lines. In that case, I believe that only deep sequencing will assure you and others that work carried on with those cell lines will not be affected by unmapped mutations.

You could always blast your guide sequence to identify potential off side targets. Then you could sequence only the top hit regions from the blast search.

Hi Manuel,

I suppose you might have been used the first generation of CRISPR/Cas9, which was shown to give a lot of off-target effects. There are resources (such as addgene and the Zhang lab) to help to predict potential off-targets in whole genome. One possibility you can do is to check these potential off-target sites (predicted from the survey) and then you will see in what percentage the off-target sites were also targeted by your CRISPR/Cas9.

Another possibility is to make new stable cell lines using the D10A mutant of Cas9, which only cuts single-stranded DNA. In this case, you need to send two constructs into the same cell to be able to generate a double-stranded break for the knockout. But this is unfortunately more time-consuming.

I can see that the community may shift to NGS related to verifying all the targeting alleles down the road. However, first we need to see substantial proof of off-targeting that remains undetected with current approaches, e.g., bioinformatics followed by experimental means such as Surveyor, ...

On top of that, chemical mutagenesis has obviously off-targeting as well, and those can not be controlled or predicted at all. And those style of experiments have been used over decades without any issues related to publishing. The advantage of chemical mutagenesis is that this method is totally unbiased.

Hence, in the mean while, perhaps the community should follow what is generally done in chemical mutagenesis style experiments: Complementation/non-complementation as well as rescue should give you a good idea about specificity.

1) The generation of two independent alleles that can be crossed together to analyze the mutant phenotype is a must. This non-complementation generally neutralizes the issues that off-target issues are hopefully not generated reproducibly. Two independent alleles can be two independent targeting events with the same targeting strategy (i.e., nuclease and template).

2) Transgenic rescue (using animals) or plasmid rescue (cells) with a WT fragment is a necessity. This is the complementation part.

To add to my comment from above.

3) If off-targeting events occurred in your two independent events, you should see different phenotypes when you compare both homozygous situations with the heterozygous situation. In this case, there may be complementation as well related to the phenotype being investigated. And a new phenotype may be exposed due to the second site hits/off-targeting events.

One neat way might be to REPAIR the original mutation that you introduced with the CRISPR/Cas9 - using homology-directed repair with a donor template in combination with the same CRISPR/Cas9 - to see if it recovers the wild-type phenotype. Off-target effects would seem very unlikely to be correctly repaired by this mechanism.

The suggestion of David is a really cool way for cell culture experiments where allelic complementation is a bit more problematic. This has been done with transposons in flies, and called a "precise" excision, where the transposon excises and leaves the WT chromosome behind. This is different compared to "imprecise" excisions where the excising transposon generates additional alleles.

Here the "precise" allele would be generated by "precise" repair of the initial mutation.

You may try to use the upcoming off-targets prediction tools ( addgene) to assesse how much off-targes in your mutant cell lines, see if it can get editor 's approval.

In addition, you can also try to use new generation of CRISPR/Cas technologies consisting of CRISPR parired nickases , which would further minimize off-target double - stranded breaks.

I agree with David - repair is the neatest, and possibly the most definitive way to address the criticism. Essentially akin to the need to complement once you have generated a knockout. Alternatively, a panel of CRISPR/Cas9 constructs targeting the same locus at slightly different positions to generate the same knockout (?) might be an option - the likelihood of distinct targeting constructs hitting the same on target AND identical off targets would be extremely unlikely, giving you further confidence that the phenotypes you're observing and attributing to the gene are real. 

Dear Manuel: I think what David Prole has said in his answer is a very near precise answer (if not the precise answer) to your problem.

Dear All,

First of all, thank you very much for the constructive & ongoing discussion. I have read with great interest all answers.

By reading through the comments, I realized that I have given too little information as to what exactly I have been doing.  Here's more details:

I have used wild-type Cas9 (Addgene’s pX458, 3rd generation Cas9 construct) plus a gRNA to induce a specific cut within an intronic region in the GOI. Simultaneously, I hit the same cells with a donor rAAV ssDNA template to induce a specific HDR at the gRNA’s cutting loci, the donor rAAV ssDNA template straddles the Cas9 cleavage site. The gRNA region has either been mutated or deleted within the donor template to avoid uncontrolled recombination events. The gRNA’s itself, have no off-target prediction sites within coding regions, following the criteria of 1) no adjacent PAM sequence, 2) 4 or more total mismatches, but 2 proximal to the PAM, 3) a total guide length of 18-19 nt, including a 5’G as part of the homology, not avoiding off-targets but in my view, dramatically reducing phenotypic off-targets. And, given the fact that a single nucleotide mismatch within the first 10-11 nt of the gRNA is most-likely sufficient to avoid the cut, the likelihood of getting a phenotype by combining all safety measures, goes to undetectable (published and my opinion). Addgene’s CRISPR Guide summarizes the chance of off-targets for this strategy as: ‘off-target effects, is not a big concern’. Still, I understand the concern.

Next, I have to explain the donor template, as I believe it allows for the selective identification of phenotypic off-target effects.

The template consists of a left and right homology arm, spanning a selection cassette; at least one of the arms contains a coding region with the desired mutation. Additionally, I introduced silent-point mutations into the coding region at another location, spanning a guide sequence of a previously well-characterized siRNA (siWT). If the recombination event occurs correctly, one allele will be wild-type (un-targeted) the other will contain: 1) the desired point-mutation and 2) the silent-point mutations that converts the mutated allele to siRNA-resistant. By adding the same silent-point mutations into the siWT siRNA sequence, I generated a new siRNA (siSN) that does not recognize the wild type mRNA sequence. Transfection of the recombined cells with either siWT or siSN siRNAs, selectively regulates the presence of either the wild-type or mutated mRNA and thereby the presence/absence of the wild-type/mutated protein, resulting in wild-type/point-mutation phenotype of the targeted protein.

In my view, if a non-specific, off-target event occurred, the resulting phenotype (if any) should be independent of wild-type or mutated protein. Hence, I believe this strategy not only allows for endogenous gene-function analysis, but for the phenotypic identification of prominent ‘off-target’ effects.

Finally, I am happy to tell you that this work has just been accepted for publication in NAR. Here the link: http://www.ncbi.nlm.nih.gov/pubmed/25586224.

Again, thank you for your comments and please let me know what you think.

Best...