Multiple-gene targeting and mismatch tolerance can confound analysis of genome-wide pooled CRISPR screens

Abstract

BackgroundGenome-wide loss-of-function screens using the CRISPR/Cas9 system allow the efficient discovery of cancer cell vulnerabilities. While several studies have focused on correcting for DNA cleavage toxicity biases associated with copy number alterations, the effects of sgRNAs co-targeting multiple genomic loci in CRISPR screens have not been discussed.ResultsIn this work, we analyze CRISPR essentiality screen data from 391 cancer cell lines to characterize biases induced by multi-target sgRNAs. We investigate two types of multi-targets: on-targets predicted through perfect sequence complementarity and off-targets predicted through sequence complementarity with up to two nucleotide mismatches. We find that the number of on-targets and off-targets both increase sgRNA activity in a cell line-specific manner and that existing additive models of gene knockout effects fail at capturing genetic interactions that may occur between co-targeted genes. We use synthetic lethality between paralog genes to show that genetic interactions can introduce biases in essentiality scores estimated from multi-target sgRNAs. We further show that single-mismatch tolerant sgRNAs can confound the analysis of gene essentiality and lead to incorrect co-essentiality functional networks. Lastly, we also find that single nucleotide polymorphisms located in protospacer regions can impair on-target activity as a result of mismatch tolerance.ConclusionWe show the impact of multi-target effects on estimating cancer cell dependencies and the impact of off-target effects caused by mismatch tolerance in sgRNA-DNA binding.Electronic supplementary materialThe online version of this article (10.1186/s13059-019-1621-7) contains supplementary material, which is available to authorized users.

Paralog dependency indirectly affects the robustness of human cells

“…Although this effect may have biased our analyses, the use of the dataset CS2, which has been corrected for copy-number variation across the cell-line genomes (Meyers et al, 2017), shows that the results are likely robust to these effects. Another factor that may affect the results is that gRNAs could potentially inactivate both paralogs of a pair, thereby leading to double gene LOF rather than a single one (Fortin et al, 2019). For instance, the CS1 original dataset (Wang et al, 2015) contained gRNAs that targeted multiple positions in the genomes, many of which could be positions that correspond to duplicated genes.…”

“…For instance, the CS1 original dataset (Wang et al, 2015) contained gRNAs that targeted multiple positions in the genomes, many of which could be positions that correspond to duplicated genes. This could lead to double gene LOF by Cas9 cutting and DNA repair but could also lead to chromosomal rearrangements, leading to even stronger effects (Després et al, 2018;Kosicki et al, 2018) than double gene LOF (Fortin et al, 2019). For this reason, we re-analyzed all data and considered only uniquely aligned gene-specific gRNAs (see Materials and Methods).…”

Extracting Biological Insights from the Project Achilles Genome-Scale CRISPR Screens in Cancer Cell Lines

“…This assumption can fail for synergistic cases, as when a single sgRNA targets many members of a family 11 . It can also introduce problems in the case of compound genes (those containing the superset of two other genes' transcripts).…”

Paralog buffering contributes to the variable essentiality of genes in cancer cell lines

“…These screens were performed using the Avana single guide RNA (sgRNA) library which contains sgRNAs targeting 17,634 genes with approximately 4 sgRNAs per gene. By design, each sgRNA should target only a single gene, but recent work has shown that this and other libraries contain 'multi-targeting' sgRNAs that can match multiple sites in the genome and consequently disrupt the function of multiple genes [24,25]. This means that any fitness measurements made using these 'multi-targeting' sgRNAs may reflect genetic interactions resulting from inhibiting multiple genes simultaneously.…”

“…This means that any fitness measurements made using these 'multi-targeting' sgRNAs may reflect genetic interactions resulting from inhibiting multiple genes simultaneously. Previous work has found that such multitargeting sgRNAs disproportionately target paralog genes [24] potentially confounding any naive comparison of the fitness effects of disrupting paralogs and singletons. To avoid such confounding we reprocessed the screens to remove multi-targeting sgRNAs ( Fig 1A).…”