Completed on 29 Apr 2016
Login to endorse this review.
Reviewer 2 Advance Summary and Potential Significance to Field:
Reviewer 2 Comments for the Author:
I have a number of comments/questions/concerns about this manuscript.
Electroporation of sgRNA drivers. The authors report that pools of electroporated embryos were used for further analysis. Does electroporation efficiency vary significantly? How many replicates per sgRNA were performed – it seems like just a single electroporation was reported for each sgRNA. What proportion of the resulting embryos were transgenic (aren’t DNA constructs mosaic ally expressed in Ciona)? Were all of these electroporations from a single batch of embryos? If not, is there variation from batch to batch? If there is variation, how does this effect your downstream analysis?
Mutagenesis frequency (pg. 6). This is really the mutagenesis frequency per haploid allele. The authors should describe how this correlates to a mutagenesis frequency per embryo, given that not all embryos express the constructs (electroporation efficiency) and not all cells within a given transgenic embryo express the transgenes (mosaic transgene expression). Presumably, there is a distribution of mutations in either allele as well as mutations in both copies of an allele within any given cell. Is there any information on what this distribution looks like?
Off-target effects (pg 7). This seems more like a specificity assay rather than an off-target assay. Mismatches outside of the PAM-proximal sequence can be tolerated and have been shown to produce DSBs – do any of your predicted sgRNAs share similar or identical PAM-proximal sequences? Have you tested DSBs on any genomic regions that have well-conserved PAM-proximal sequences, but have mismatches outside of this region? Did you consider single nt mismatches to the PAM proximal sequence, rather than just 2 or 4 nt differences? (2 or 4 nt mismatches are not likely to be effective in any case). This would be a more accurate assessment of off target cleavage. One of the papers you reference, Hsu et al., assay substantially more potential off-target genomic locations; it would be prudent to include a much larger number of genomic sites to assay off-target effects. It would be useful to include a supplementary table that describes these potential off-target sites and the resulting analysis of such sites.
On page 9 you mention that CRISPRScan is a tool for rational sgRNA based on zebrafish data. You developed your own algorithm (TuniCUT) because you hypothesized that there would be differences between Ciona and zebrafish. However, you never compare Ciona sgRNAs designed with CRISPRScan to those designed with TuniCUT. Do they differ substantially? If not, then what is the significance of your algorithm? If they are similar, what does that say about the mechanism of Cas9 activity in Ciona vs. other species? If there is not a significant difference, then the AT-rich nature of the Ciona genome has little to no effect on the CRISPR/Cas9 process, only on the ability to locate a suitable CRISPR target sequence.
Is the training set large enough to provide sufficient discrimination of “good” and “bad” sites for your algorithm? CRISPRScan used >1200 sgRNAs; you essentially analyzed < 50 sgRNAs (~20 good and ~20 bad – 25% of the 83 total sgRNAs). It would seem that this small number of analyzed sites could significantly skew your results. Again, with no comparisons to CRISPRScan, it is unclear if your algorithm provides any advantages over existing sgRNA design algorithms.
On page 9 you mention that you added an arbitrary error of 10% based on plasmid DNA uptake. What exactly does this mean, and how did you settle on a value of 10%?
It is not clear that you have experimentally demonstrated that your algorithm can accurately design functional sgRNAs as it seems most of the sgRNAs you report in this study were the same ones used as inputs to your algorithm. If this is not the case, then this was not clear from the text. I would have expected that you would test your algorithm by comparing say one or two dozen novel sgRNAs predicted to work well vs. one or two dozen sgRNAs that have much lower expected function. This comparison would at least provide experimental evidence to support your scoring scheme. Ideally, this should also be compared with sgRNAs identified by CRISPRScan to assess whether your algorithm is a better predictor of functional sgRNAs in Ciona.
On pages 10-11, you report on producing large deletions by using multiplexed sgRNAs. However, your assay only detects the presence of a deletion product. Do you know what percentage of alleles are deleted within an embryo? Is this a rare event? Have you analyzed/quantitated the spatial distribution of these events?
Can you compare the relative amounts of “wild-type” to deleted regions as a more accurate measure of efficiency? In other words, compare short PCR products from your specific deleted region to similarly-sized PCR products produced from non-deleted regions. This should provide a more accurate, quantitative description of your mutation efficiencies.
On page 12 you describe the use of linear PCR products to express sgRNAs. It is a common practice when generating stable cell lines to linearize a plasmid before transfection to increase the probability of genomic integration.
Supercoiled plasmids are much less efficiently integrated. Do you know if linear DNA/PCR products integrate into the Ciona genome? What about supercoiled
plasmids? Is this a cause of concern – the potential introduction of additional mutations due to the random integration of linear DNAs?
Secondly, you only report a single sgRNA introduced as a PCR product (Ebf.3). Do you know if this works for a wide variety of sgRNAs? Do the linear products work better than the corresponding plasmid product? Do you know how much sgRNA product is produced from the PCR product vs. the plasmid form? Do pools of PCR products work? It seems that there is far less certainty about the usefulness of PCR products than plasmid products.
Supplemental protocol (page 50). In your supplementary protocol, you explain that your pre-designed sgRNAs must first be checked with CRISPRdirect to identify off-targets and then you must check for polymorphisms with the Ciona genome browser. Shouldn’t your design algorithm already include this information for the end user?