To add to the growing knowledge base for gRNA design, Wong et al (2015) revisit the data generated by Doench et al (2014), which employed a lentivirus library to screen >1800 sgRNA for activity in mouse EL4 cells.
The authors chose an approach that has been successful in the design of highly active dsRNAs for RNA interference and compared the top 20% of guides (those with the highest cutting activity) with the bottom 20%.
Factors that they took into consideration were sequence and structural characteristics (namely nucleotide accessibility), overall nucleotide usage, and the presence of certain nucleotides at defined positions.
Similar to previous findings, the results demonstrated that nucleotides in the 3’-PAM proximal region of the protospacer were important, which the authors attributed to nucleotide accessibility (with respect to the secondary structure of the sgRNA as predicted by RNAfold in the Vienna package). In addition to the 3’-PAM proximal region of the protospacer, they also concluded that nucleotide positions 51-53 (occurring in the scaffold region of the sgRNA) were important and more “accessible” in highly active sgRNAs. Further structural analyses revealed distinct profiles in the self folding energy and duplex binding stability for highly active sgRNA compared to the bottom 20%.
The WU-CRISPR algorithm is available as an online tool for checking the predicted activity of a gRNA at http://crispr.wustl.edu/ . Models are available for human and mouse, or sequences can be directly copied and pasted, however they will not search for off-targets or be able to provide genomic primer sets for downstream analyses. Yet, if you are interested in obtaining the WU-CRISPR score for a guide you have already identified, you only need to paste the protospacer sequence (at least 20 nt long) followed by the PAM+1 = 24 nt total.
When using WU-CRISPR, it is important to keep in mind some features of the algorithm. A major consideration is that the default thresholds for efficacy are quite stringent, which results in the null prediction of some guides, since the minimum reporting score threshold is 50 out of 100. In this regard, users may find the stand-alone perl script to be less strict, since lines can be modified within the script to remove some of these filters, allowing for low-scoring guides to be predicted.
Finally, it may be good to keep in mind that the data set utilized to generate the algorithm for WU-CRISPR is from mouse cell culture and PolIII driven sgRNA. If your experimental approach is different, the activity prediction from WU-CRISPR may not be meaningful.
For example, Basu et al (2015), report on the experimentally-determined activity of 52 sgRNA tested in Ae. aegypti by injecting either Cas9 protein or mRNA along with the in vitro synthesized sgRNAs. Of these sgRNA, 18 were non-active while 34 were active, with several sgRNA of noteworthy high activity.
When the guides from Basu et al were predicted for activity using the online version of WU-CRISPR, four guides from the 18 non-functional guides were predicted as having activity ≥ 50, scoring from 53 to 71. From the 32 active guides, 13 were predicted to be functional, scoring from 50 to 87.
When the stand-alone version was downloaded, and the “activity ≥ 50 minimum reporting threshold” was removed, a total of 24 of the 52 guides could be assessed for their WU-CRISPR activity.
Overall, a poor correlation was found for the experimentally determined measures of in vivo gRNA activity for the gRNA from Basu et al (2015) and the WU-CRISPR algorithm (Figure). Thus while WU-CRISPR is another predictive tool for deciding which guides to select for your next CRISPR experiment, it may miss guides that have genuinely good cutting activity, and may also predict guides with poor activity as being “highly active”.
It is clear that predicting ‘good’ guide RNAs remains challenging and one should hedge against being misled by existing algorithms by incorporating some kind of initial in vivo validation step into your mutagenesis scheme.
Nathan Wong, Weijun Liu, Xiaowei Wang (2015) WU-CRISPR: characteristics of functional guide RNAs for the CRISPR/Cas9 system. bioRχiv