CRISPR-Cas9 screening

The CRISPR-Cas9 screen was performed using the Broad Institute’s GeCKO library (Sanjana et al., 2014; Aguirre et al., 2016). 33 cancer cell lines (including nine Ewing sarcoma lines) were screened with the GeCKO library, containing ∼95,000 guides and an average of six guides per gene (Sanjana et al., 2014; Aguirre et al., 2016). The library contains ∼1,000 negative control guides that do not target any location in the reference genome. The library also included guides with more than one perfect match in the reference genome allowing us to computationally correct for the previously described cutting toxicity associated with multiple Cas9 cuts in the genome (Aguirre et al., 2016).

Cancer cell lines were transduced with Cas9 using a lentiviral system (Aguirre et al., 2016). Cell lines that met quality control criteria, including Cas9 activity measured using a GFP reporter, and other parameters, were then screened with the CRISPR library. A pool of guides was transduced into a population of cells. The cells were cultured for ∼21 d in vitro, and at the end of the assay, barcodes for each guide were sequenced for each cell line in replicate. Reads per kilobase were calculated for each replicate and then the log2 fold change compared with the initial plasmid pool was calculated for each guide. Samples with poor replicate reproducibility, as well as guides that have low representation in the initial plasmid pool, were removed from the analysis. Next, the guides from multiple replicates for each sample were used to collapse the data into gene scores using the CERES algorithm (Meyers et al., 2017), which models the cutting effect of each guide correcting for multiple cuts in the genome to produce a score that reflects the effect of disruption of the gene. After the dependency scores were calculated using the CERES algorithm, the scores for each cell line were scaled so that mean of negative controls was 0 and the mean of a subset of positive controls was −1.

For all of our analysis, the data were filtered and only the set of genetic dependencies with variable dependency scores that had standard deviations two sigma above the mean standard deviation across all genes were used. This resulted in 705 dependencies. Pearson correlations were then computed between the dependency gene score for TP53 and all other variable dependencies in the screen. The top eight anti-correlated genes were used for subsequent analysis.