A compendium of synthetic lethal gene pairs defined by extensive combinatorial pan-cancer CRISPR screening

Abstract

Background: Synthetic lethal interactions are attractive therapeutic candidates as they enable selective targeting of cancer cells in which somatic alterations have disrupted one member of a synthetic lethal gene pair while leaving normal tissues untouched, thus minimising off-target toxicity. Despite this potential, the number of well-established and validated synthetic lethal gene pairs is modest.

Results: We generate a dual-guide CRISPR/Cas9 Library and analyse 472 predicted synthetic lethal pairs in 27 cancer cell Lines from melanoma, pancreatic and lung cancer Lineages. We report a robust collection of 117 genetic interactions within and across cancer types and explore their candidacy as therapeutic targets. We show that SLC25A28 is an attractive target since its synthetic lethal paralog partner SLC25A37 is homozygously deleted pan-cancer. We generate knockout mice for Slc25a28 revealing that, except for cataracts in some mice, these animals are normal; suggesting inhibition of SLC25A28 is unlikely to be associated with profound toxicity.

Conclusions: We provide and validate an extensive collection of synthetic lethal interactions across cancer types.

Keywords: CRISPR; Epistasis; High-throughput screening; Synthetic lethality.

Fig. 1

Design and implementation of a dual guide CRISPR library to detect synthetic lethal gene pairs. A Schematic of the dual-guide vector used for library construction containing a human U6 (hU6) and a mouse U6 (mU6), two gRNAs and two tracrRNA sequences in a lentiviral backbone (see Methods). B The Library contained a total of 22,823 unique gRNAs, targeting 472 gene pairs selected from paralogous genes, MASH-up analysis of Cancer Genome Atlas (TCGA)/Achilles data, and gene pairs derived from an analysis to associate gene expression and CRISPR essentiality (see Methods). Controls included essential/non-essential genes and safe-targeting controls. Each of the abovementioned gRNAs targeting gene pairs were paired with one of the safe-targeting control sequences in the library allowing us to compute the cell fitness effect of single gene disruption. C Outlines the methodology used to select the gene pairs analysed in this study. Linear regression associations between CRISPR-Cas9 and RNA-Seq datasets were statistically assessed using log-ratio tests, and multiple hypothesis testing correction was applied using the Benjamini–Hochberg False Discovery Rate (FDR) method

Fig. 2

Identification of synthetic lethal gene pairs across 27 cancer cell lines. Heatmap depicting all gene pairs that were classed as synthetic lethal (SL) in 3 or more cell lines. Box colour indicates the mean normalized GI score with pairs considered a hit if the mean normalised GI score was < −0.5 and the Benjamini–Hochberg False Discovery Rate (FDR) < 0.01 (normalised GI scores mean comparison using T-test). Blanks indicate where a gene pair was not a hit in that cell line. Black vertical lines are used to separate cell lines from lung, melanoma, and pancreas. The cell line names are shown on the X-axis. Right: Bar plot shows the frequency that the indicated pairs were a hit with the bar coloured by the cancer type

Fig. 3

Overlap of our hits and those identified by other published dual guide CRISPR screens. A We compared the data from the screens in this paper to those from the Synthetic Lethality Knowledge Database (SLKB) [34]. The screens performed in this study are shown on the right of the body, with those in SLKB shown on the left. Commonly screened cell lines are highlighted in red. The number in brackets next to the cell line name refers to the number of independent screens performed in that line. The anatomical locations are approximate and for reproductive system cancers the sex of the cell lines is shown (M; male. F; female). B Shown are diagrams reflecting the overlap of data from the 472 gene pairs screened in this study. We identified 117 screen “hits” across all 27 cell Lines screened. Of these 472 pairs, 272 were found in SLKB and 130 of these pairs were not called as synthetic lethal in our screens or in any SLKB screen (neg pairs). Shown below (box diagram) is a breakdown of the 117 screen hits in comparison SLKB data. The full dataset is available in Additional File 1: Table S6

Fig. 4

Hits are context specific and not generally associated with cancer type. A The mean normalised GI score for each cell line in which the gene pair was classified as a hit is shown. Pairs were considered a hit if the mean normalised GI score was < −0.5 and the Benjamini–Hochberg False Discovery Rate (FDR) < 0.01 (normalised GI scores mean comparison using T-test). Score for each cell line is coloured by cancer type. B Pie chart indicating the breakdown of the 472 pairs into never a hit (not a hit in any line), context dependent hit (hit in less than 50% of our cell lines), or strong hit (hit in more than 50% of our cell lines). C Pie chart showing the number of lines in which each gene pair was classed as SL (excluding those gene pairs that were not a hit in any line). D Overlap of hits by cell line cancer type. This Figure depicts gene pairs that were a hit in one or more cell line

Fig. 5

Using GI score and gene pair rankings to define the strongest context independent SL gene pairs. A Median GI score per gene pair in which the pair was classed as a hit, plotted against the number of cell lines the gene pair was a hit. Line shows the linear regression analysis with the Pearson’s r correlation coefficient and p-value computed to be −0.557987 and p-value = 6.326e-11, respectively. B Range of mean normalised GI scores for hits per cell line. C Per cell line each hit was ranked by GI score (lowest GI score for each Line ranked as 1). Graph shows the ranking of each hit per line for the gene pairs classed as strong hits (i.e. a hit in more than 50% of screened lines). D The effect of cell lineage on synthetic lethality/GI score. P values are derived from a Type II ANOVA analysis with Benjamini–Hochberg correction. E–F Expression impacts the penetrance of genetic interactions. LFC represents the effect if single gene and double gene disruption. Scaled mRNA levels are shown. GI refers to the GI score

Fig. 6

Phenotypic validation of strongest hits by siRNA knockdown and subsequent cell imaging. A GI scores in the A549 cell line for the pairs analyses by imaging. B Percentage of cells defined as apoptotic, enlarged, non-proliferative and proliferative. C Fold change in apoptotic, enlarged, non-proliferative and proliferative for genes and gene pairs analysed in this experiment. The data represent the median of four independent experiments with the lines and bars the IQRs. D Representative pictures of cellular phenotypes observed following gene co-disruption. Colours are as follows: DNA (blue), tubulin/spindles (green), Annexin V/apoptosis (red)

Fig. 7

Positioning of synthetic lethal interactions via comparison to human cancer and normal tissue datasets. A Analysis of pan-cancer copy number profiles for selected synthetic lethal gene pairs with the data expressed as a percentage frequency for each TCGA cancer type. The data was downloaded from the Xena Browser [45] and has been described previously [46]. Dark blue denotes homozygous loss of each gene by cancer type. B Expression of synthetic lethal genes across cancer types. Shown are the transcripts per million values. C Expression of synthetic lethal genes across normal tissues. Shown are the transcripts per million values. The data was downloaded from GTEx [38]. D We identified 62 gene pairs for which we observed ubiquitous expression in normal tissues, loss of expression of one member of the pair in tumours (left), where the pair was also found to be a screen hit. Expression was defined as > 1 TPM