Genome-Wide CRISPR Screen Identifies KEAP1 Perturbation as a Vulnerability of ARID1A-Deficient Cells

Abstract

ARID1A is the core DNA-binding subunit of the BAF chromatin remodeling complex and is mutated in about 8% of all cancers. The frequency of ARID1A loss varies between cancer subtypes, with clear cell ovarian carcinoma (CCOC) presenting the highest incidence at > 50% of cases. Despite a growing understanding of the consequences of ARID1A loss in cancer, there remains limited targeted therapeutic options for ARID1A-deficient cancers. Using a genome-wide CRISPR screening approach, we identify KEAP1 as a genetic dependency of ARID1A in CCOC. Depletion or chemical perturbation of KEAP1 results in selective growth inhibition of ARID1A-KO cell lines and edited primary endometrial epithelial cells. While we confirm that KEAP1-NRF2 signalling is dysregulated in ARID1A-KO cells, we suggest that this synthetic lethality is not due to aberrant NRF2 signalling. Rather, we find that KEAP1 perturbation exacerbates genome instability phenotypes associated with ARID1A deficiency. Together, our findings identify a potentially novel synthetic lethal interaction of ARID1A-deficient cells.

Keywords: ARID1A; CRISPR screening; KEAP1; synthetic lethality.

Figure 3

ARID1A-KO sensitivity to KEAP1 depletion is only partly due to dysregulation of NRF2 (A) Volcano plot of differentially expressed genes in ARID1A-KO RMG-1 cells from RNA-seq dataset published in Wu et al. [22] (blue dots if padj < 0.01, red dots if log2FC > 1 or log2FC < −1 AND padj < 0.01). (B) Top 10 enriched transcription factor target genes identified from the ARID1A-UP (above) and ARID1A-DOWN (below) regulated genes using the ENCODE [50] and ChEA [51] datasets (EnrichR [52]; p < 0.05 with Benjamini-Hochberg correction). (C) Quantification of KEAP1 and ARID1A mRNA levels (RSEM, log2) from pan-cancer TCGA data (n = 10,071) showing a negative correlation between the two genes. (D) Quantification of KEAP1 mRNA levels (RSEM, log2) from TCGA data (stomach adenocarcinoma—STAD) showing higher KEAP1 levels in ARID1A-mutated patient samples (n = 90) compared to WT (n = 336) (box limits indicate 25% and 75% over median, Wilcoxon test, p = 2.8 × 10⁻⁶). (E) Left: Quantification of KEAP1 protein levels from western blot data showing higher levels in ARID1A-KO RMG-1 cells. KEAP1 intensity was normalized to respective GAPDH loading control, and then to ARID1A-WT (data from 9 from independent samples, mean ± SEM, t test, significant p-values displayed on graph). Right: Representative western blot image. (F) Quantification of crystal violet viability assay showing that ARID1A-KO RMG-1 cells exhibit sensitivity to NRF2^E79Q expression induced by doxycycline (mean ± SEM, ANOVA, n = 3, significant p-values displayed on graph).

Figure 1

Genome-wide CRISPR screen identifies KEAP1 as a synthetic lethal partner of ARID1A. (A) Schematic of CRISPR screening workflow with the TKOv3 library. (B) Precision-Recall curve assessing CRISPR screen performance of individual replicates (values obtained from BAGEL2 algorithm, Supplementary Table S1.5-6). (C) Top 10 enriched Biological Processes GO terms identified using David v 6.8 (p < 0.05, with FDR correction) for the ARID1A-KO fitness genes identified by either of our CRISPR screens (n = 1484 genes, Supplementary Table S1.7). (D) Overlap of ARID1A-SL hits from both biological replicates of the CRISPR screen. (E) Ranked ARID1A-KO specific hits (average BF from both biological replicates, Supplementary Table S1.8). The top 3 hits are labeled on the graph. Panel (A) was created using Biorender.

Figure 2

ARID1A-KO cells are sensitive to KEAP1 depletion by siRNA and small molecule perturbation. (A) Quantification of crystal violet viability assays in RMG-1 cells (mean ± SEM, one-way ANOVA, n = 3, statistically significant p-values are displayed on graph). (B) ARID1A-KO RMG-1 cells grow slower than ARID1A-WT when treated with 75 µM AI-1 as measured by IncuCyte S3 imaging system. Relative confluency presented as the growth from initial time point. Error bars represent SEM of averaged triplicated wells from 3 independent experiments. P values obtained from extra sum-of-squares F test on calculated logistic growth rate are indicated on graph. (C) ARID1A-KO CCOC cells present lower IC50 values to AI-1 than ARID1-WT cells (t-test, mean ± SEM). IC50 values were obtained from dose-response curves presented in Supplementary Figure S2F–H (n = 3). (D) Patient-derived endometrium progenitor cell workflow. Normal endometrial tissue was processed and put in 2D culture to expand the progenitor cell population before CRISPR and AI-1 treatments (see Section 4). (E) Average proliferation curves of three primary endometrial epithelial cell cultures transduced with non-targeting (NTC5) or sgARID1A (ARID1A-KO) lentivirus and treated with AI-1 at the indicated concentrations. Technical quadruplicates from each patient were normalized and combined. p-value was determined at the midpoint for 60 µM AI-1 (Welch’s t-test). CRISPR knockdown western blots, and individual patient derived growth curves are shown in Supplementary Figure S2E–H. Panel (D) was created using Biorender.

Figure 4

KEAP1 perturbation exacerbates genome instability phenotypes of ARID1A-KO cells (A,B). Quantification (left) and representative images (right) of immunofluorescence experiment showing AI-1 (50 µM) induces higher rates of p-RPAser33 (A) or γH2AX (B) foci formation in ARID1A-KO RMG-1 cells (mean ± SEM, ANOVA, n = 3, significant p-values displayed on graph). (C,D) Quantification (left) and representative images (right) showing that ARID1A-KO RMG-1 cells accumulate higher levels of micronuclei compared to WT when treated with 50 µM AI-1 (C) or siKEAP1 (D) (mean ± SEM, ANOVA, n = 3, significant p-values displayed on graph). White arrows highlighting micronuclei. In A through D, a 20 µm scale bar is show in the lower right.

Figure 5

Sensitivity of ARID1A-KO cells to KEAP1 perturbation is conserved in endometrium progenitor-derived organoid model. Model of ARID1A-KEAP1 synthetic lethality. Perturbation of KEAP1 dysregulates normal NRF2 signalling and proteostasis, which remains viable in normal cells. Perturbation of KEAP1 in ARID1A-KO cells specifically enhances genome instability phenotypes, resulting in growth defects. This figure was created using Biorender.