Genome-wide CRISPR/Cas9 screens are a high-throughput systematic approach for identifying genes involved in a biological process. These screens provide an alternative to genome-wide RNAi screens, which although highly effective, are affected by low on-target efficacy, non-specific toxicity, and off-target effects. The flaws of RNAi screens are well characterized and strategies exist to control for these faults. However, it’s still unclear if similar pitfalls exist for CRISPR screens and how best to design these screens to controls for flaws. Recently the Bassik Lab at Stanford developed a new genome-wide CRISPR knockout screen to analyze the following unanswered questions about CRISPR screen design.
Unanswered questions about genome-wide CRISPR/Cas9 screen design
- Are non-targeting guides an appropriate control? Many CRISPR screens use non-targeting guides as negative controls. Non-targeting guides do not target any site in the genome. This means that while they control for many of the effects of gRNA and Cas9 expression, they fail to account for the effects of Cas9-induced dsDNA breaks.
- Are off-target cutting patterns in genome-wide screens similar to patterns seen in single-guide studies? Many strategies exist for reducing Cas9 off-target effects, but most studies have focused on the off-target cutting of only a handful of guides. It’s unclear if off-target cuts confound the results of high-throughput CRISPR screens.
- Do short guides (17-18 bp) have less off-target cutting than full-length guides (19-20bp)? Small-scale studies (Tsai et al., Fu et al.) suggest 17-19 bp long guides have reduced off-target cutting without a reduction of on-target activity, but it’s unclear if this holds true in genome-wide screens.
Bassik lab’s CRISPR knockout libraries
Overview of Library
To better define the optimal conditions for CRISPR knockout screens, the Bassik lab created a human knockout library that has 10 gRNAs per gene and targets all ~20,500 protein-coding genes. Unique, non-overlapping sites in the genome were targeted and predicted on-target activity was balanced with predicted off-target activity when selecting guides. Guides were delivered as a pooled library via a lentiviral vector to cell lines that either stably expressed Cas9 or that had been lentiviral infected with Cas9.
Validating the Library
The CRISPR knockout library was validated via a growth screen and a ricin toxicity screen since essential and non-essential genes for both pathways had previously been identified (Hart et al, Bassik et al). The library performed well in both screens, with the growth screen having a 1% false discovery rate and identifying >88% of previously identified essential genes while previous Cas9 and shRNA library screens only identified 60% of the essential genes. The ricin toxicity screen had a 10% false discovery rate and identified 67% of previously identified ricin toxicity genes. The screen also discovered several genes previously not associated with ricin regulation, including almost all genes involved in the production of a cell surface glycan that’s required for ricin uptake. The ricin screen did fail to identify some known ricin-regulators, but most of these genes are also essential for growth and would not be expected to be identified in a CRISPR knockout screen.
Improving the Design of CRISPR Knockout Screens
Before diving into the results, it’s important to note that in Morgens et al., guides were considered toxic if they were depleted from the screen. This toxicity was used a proxy for detecting Cas9 cutting.
Safe targeting guides better control for non-specific toxicity than non-targeting guides
First, Morgens et al. looked at the effect of safe-targeting guides vs. non-targeting guides on screen analysis and hit calling. 5,644 non-targeting guides and 6,750 safe-targeting guides were included in the library. Safe-targeting guides were designed to target genomic sites with no annotated function, i.e. sites that lack open chromatin marks, DNase hypersensitivity, or are in an enhancer, transcription factor binding site, or a coding region. Targeting these sites should control for the effects of guide expression and dsDNA breaks.
In the growth screen, safe-targeting guides were depleted at greater rates than their non-targeting counterparts, suggesting that safe-targeting guides are more toxic than non-targeting guides. Toxicity is likely due to DNA damage and its subsequent repair. Additionally, the distribution of enrichment and depletion scores for safe-targeting guides more closely mirrors that of gene-targeting guides, suggesting that safe-targeting guides better control for the inherent effects, or background noise, of dsDNA breaks caused by CRISPR. See supplemental figure 6 for more details.
Safe-targeting guides also alters the discovery of hits from the growth and ricin screen. When safe-targeting guides are used to call hits from the growth screen, p-values were less significant than when non-targeting guides are used as controls. Less significant p-values lead to more false-negatives results (i.e. failing to identify a gene that’s a hit); but they also lead to fewer false-positives results (i.e. mistakenly calling a gene a hit when really it’s not). In practical terms, this means that using safe-targeting guides as a control will result in a smaller list of true positive hits than when using non-targeting guides, when both are analyzed using the same false-discovery rate cutoff.
Adding safe-targeting guides to the CRISPR library resulted in both increased and decreased p-values of known ricin regulation genes compared to results generated using non-targeting guides as controls. While it’s not clear how safe-targeting guides will affect phenotype in all non-growth screens, these results suggest that safe-targeting guides may serve as better controls than non-targeting guides and may more accurately determine the significance of hits in non-growth screens.
Off-target cutting does not confound results of CRISPR screen
Next, the toxicity of guides with off-target sites was profiled. Guides with exact or 1-bp mismatch off-targets had greater toxicity than guides that had zero mismatch off-targets. Guides with 2-bp mismatch off-targets were only toxic if they had 5+ off-target sites. Results from the screen also reproduced several previously observed characteristics that influence gRNA off-target activity: 1) mismatches closer to the PAM are less tolerated than mismatches more distal to the PAM, and 2) guides with high GC content have greater off-target activity.
Short gRNAs have lower off-target cutting and similar on-targeting cutting as full-length gRNAs
Lastly, the effects of guide length on off-target cuts was explored using full-length (19-20 bp) or short (17-18 bp) guides. When comparing the toxicity of guides that have 1-bp mismatch off-target sites, short guides had reduced toxicity compared to full-length guides. Reduced toxicity of short guides could be due to lower off-target activity, but it could also be caused by a reduction of on-target activity. However, there was no significant difference in depletion of ricin sensitizer genes identified with short and long guides. Together these results suggests that short guides have fewer off-target effects than long guides, and do not have a major reduction in on-target efficacy.
Results from Morgens et al. show that genome-wide CRISPR screens are prone to some of the same flaws as RNAi screens, such as non-specific toxicity due to both on- and off-target effects, but also presents a strategy to control for these effects. By testing thousands of guides for off-target activity, Morgens et al drew some generalizable conclusions about off-target activity, demonstrate safe-targeting guides as a potential control, and provide evidence that truncated gRNAs have improved specificity with only a small loss of on-target activity in high-throughput screens. The human and mouse versions of the Bassik CRISPR Knockout Library are available from Addgene!
References
1. Morgens, David W., et al. "Genome-scale measurement of off-target activity using Cas9 toxicity in high-throughput screens." Nature communications 8 (2017): 15178. Pubmed PMID: 28474669. Pubmed Central PMCID: PMC5424143.
2. Tsai, Shengdar Q., et al. "GUIDE-seq enables genome-wide profiling of off-target cleavage by CRISPR-Cas nucleases." Nature biotechnology 33.2 (2015): 187. Pubmed PMID: 25513782. Pubmed Central PMCID: PMC4320685.
3. Fu, Yanfang, et al. "Improving CRISPR-Cas nuclease specificity using truncated guide RNAs." Nature biotechnology 32.3 (2014): 279. Pubmed PMID: 24463574. Pubmed Central PMCID: PMC3988262.
4. Hart, Traver, et al. "Measuring error rates in genomic perturbation screens: gold standards for human functional genomics." Molecular systems biology 10.7 (2014): 733. Pubmed PMID: 23394947. Pubmed Central PMCID: 4299491.
5. Bassik, Michael C., et al. "A systematic mammalian genetic interaction map reveals pathways underlying ricin susceptibility." Cell 152.4 (2013): 909-922. Pubmed PMID: 23394947. Pubmed Central PMCID: 3652613.
Additional Resources on the Addgene Blog
- Learn to conduct genome-wide screens with CRISPR/Cas9
- Review the basics of CRISPR with this cheat sheet
- Read our Quick Guide to All Things Lentivirus
Resources on Addgene.org
- Check out this guide to pooled plasmid libraries
- Find all of the CRISPR Pooled Libraries currently available from Addgene
Other Resources
- Watch Dr. Michael Bassik’s lecture on “Multiplexing with CRISPR Screens”
Topics: CRISPR, CRISPR Pooled Libraries
Leave a Comment