Deciphering how genes interact within human cells is essential for understanding their functional wiring and for developing targeted therapeutic strategies. In this study, we present a genome-scale map of genetic interactions in the human haploid cell line HAP1, based on CRISPR-based perturbation of ∼4 million gene pairs. The resulting network comprises ∼89,000 high-confidence gene-gene interactions, organizing genes into hierarchical modules corresponding to protein complexes and pathways, biological processes, and cellular compartments, mirroring principles observed in yeast and highlighting the functional architecture of a human cell. This large-scale genetic network complements the DepMap gene co-essentiality network by capturing unique functional information, uncovering roles of previously uncharacterized genes, and identifying molecular determinants of cancer-cell-line-specific genetic dependencies. This study presents a general data-driven strategy for systematically exploring the roles of genes and their functional connections in human cell lines.

Interest in understanding the high chemoresistance and poor prognosis of advanced ovarian clear cell carcinoma (OCCC) is rising. Patient-derived xenografts (PDX) are widely used in vivo models because of their supposedly accurate morphologic and (epi)genetic representation of patient tumors. Here, we established five subcutaneous OCCC PDXs. The PDX.F1 engraftment success rate was over 30% with similar latency time and growth speed of PDX.F2. ARID1A, PTEN, ATM, BRCA1 and PIK3CA mutations were found in matched tumors and PDXs. ARID1A protein loss was further verified by immunohistochemical staining. Cyclophilin A staining depicted the replacement of human stroma by mouse stroma in PDX.F2, while PAS/PAS-D staining confirmed cellular glycogen accumulation in OCCC tumors and PDXs. SNP array and Infinium MethylationEPIC BeadChip array data analysis demonstrated the copy number alterations and DNA methylation signatures of genome-wide and tumor-driver genes in PDXs generally resembled their patients’ tumors. Promoter CpG islands of a small number of genes, enriched in PRC2/histone methylation related gene-sets, gained methylation (△β-value > 0.4) in PDXs vs patient tumors. In conclusion, the high phenotypic and molecular similarity allows the established PDXs to serve as potential preclinical models for future translational research of OCCC.

We generated a genome-scale, genetic interaction network from the analysis of more than 4 million double mutants in the haploid human cell line, HAP1. The network maps ~90,000 genetic interactions, including thousands of extreme synthetic lethal and genetic suppression interactions. Genetic interaction profiles enabled assembly of a hierarchical model of cell function, including modules corresponding to protein complexes, pathways, biological processes, and cellular compartments. Comparative analyses showed that general principles of genetic networks are conserved from yeast to human cells. A genetic interaction network mapped in a single genetic background complements the DepMap gene co-essentiality network, recapitulating many of the same biological connections and also capturing unique functional information to reveal roles of uncharacterized genes and molecular determinants of specific cancer cell line genetic dependencies.

Supplementary Figure 5 | Clustering (n=106) and survival analysis (n=70) on kinome mutations and CNA. (A) Disease-specific survival of patients (n=70) grouped on ARID1A status, (B) PIK3CA status and (C) ARID1A + PIK3CA status. (D) Consensus clustering, with maximum group count set to 10 and clustering optimization with 1000 repetitions maximum, shows adding of tumors (n=106) as consensus index (horizontally) and empirical cumulative distribution (vertically) for 1-10 clusters. (E) Area under the curve plot of decrease in friction from 1-10 clusters. (F) Heatmap distribution of tumors in 8 clusters. (G) Disease-specific survival of patients (n=70) in 8 clusters, (H) cluster 3 vs. other clusters. (I) Disease-specific survival in advanced stage OCCC patients (FIGO 2C-4), Log rank (Mantel-Cox) was used for statistical analysis. (J) Nonsynonymous mutation distribution in genes involved in the frequently mutated PI3K/AKT/mTOR (blue) and MAPK pathway (yellow), ERBB family of receptor tyrosine kinases (green) and DNA repair pathway (red) as well as ARID1A and PALB2 are shown with OncoPrint. CNA in each mutated gene are added. The 106 OCCC tumors that were both kinome sequenced and SNP arrayed are shown on the horizontal axis grouped on tumor clusters and ordered on total event frequency in the subsequently altered pathways. (K) Oncoprint from advanced stage patients in cluster 3 vs. other clusters, shown on the horizontal axis grouped on tumor clusters and ordered on total event frequency in the subsequent altered pathways, ordered as aforementioned.

Supplementary Figure 4 | Whole genome CNA heatmap. (A) Genome-wide CNA heatmap profiles of all 108 OCCC tumors (above), 63 ARID1A wildtype tumors (middle) and 45 ARID1A mutant tumors (below). (B) Significant CNA plot of only ARID1A wildtype tumors (n=63) and (C) only ARID1A mutant tumors (n=45) as determined by GISTIC analysis. All kinases and cancer-related genes from the kinome sequencing gene panel that were focally significantly amplified (red) or deleted (blue) are indicated along the chromosomes vertically. Chromosomal location and total amount of tumors harboring the event are annotated with each gene name. False-discovery rate (FDR) 0.05 threshold, indicated by the green line, and G-score are shown along the horizontal axis.

Supplementary Table 3 | Binomial testing of kinome sequencing variants.

Supplementary Table 1 | Gene set for kinome sequencing.

Supplementary Figure 7 | PDX alteration status and p-S6 staining. (A) Nonsynonymous mutation distribution in PDX.155, PDX.180 and PDX.247 in genes involved in the frequently mutated PI3K/AKT/mTOR (blue) and MAPK pathway (yellow), ERBB family of receptor tyrosine kinases (green) and DNA repair pathway (red) as well as ARID1A and PALB2 are shown with OncoPrint. CNA in each mutated gene are added. (B) PDX.155, (C) PDX.180 and (D) PDX.247 representative p-S6 expression after 21 days of vehicle or AZD8055 treatment.

Supplementary Figure 2 | Significantly mutated genes AKT1, PIK3R1, ERBB3, FBXW7, ATM, CHEK2 and MYO3A. Schematics of the identified novel significantly mutated genes (A) AKT1, (B) PIK3R1, (C) ERBB3, (D) FBXW7, (E) ATM, (F) CHEK2 and (G) MYO3A in OCCC. Mutation marks are shown in black (truncating), red (SIFT and PolyPhen damaging prediction), yellow (SIFT or PolyPhen damaging prediction) or white (SIFT and PolyPhen benign prediction). Mutation effects are indicated with a black spot when paired control was available and written in black (previously described mutation) or red (novel mutations).

Supplementary Figure 3 | Mutation distribution in OCCC. (A) Nonsynonymous mutation distribution in genes involved in the frequently mutated PI3K/AKT/mTOR (blue) and MAPK pathway (yellow), ERBB family of receptor tyrosine kinases (green) and DNA repair (red) pathway is shown. In this OncoPrint, kinome sequenced OCCC tumors are depicted on the horizontal axis and ordered on mutation frequency in the subsequent altered pathways, represented vertically on the right. ARID1A mutant tumors are displayed at the top. (B) BioVenn diagrams demonstrating overlap in ARID1A mutant tumors and PI3K/AKT/mTOR, MAPK and DNA repair pathway and ERBB family of receptor tyrosine kinases mutant tumors (n=103). On the right, PI3K/AKT/mTOR and MAPK pathway, the ERBB family of receptor tyrosine kinases and DNA repair pathway mutations within only ARID1A wildtype tumors (n=49) and only ARID1A mutant tumors (n=44). All data in A and B is derived from 122 kinome-sequenced OCCC tumors, overlap in BioVenn diagram circles is proportional to group overlap.

Legends of supplementary figures and tables

Supplementary Table 4 | Clinicopathological characteristics (n=70 patients).

Supplementary Figure 6 | mTORC1, mTORC1/2 and PI3K-mTORC1/2 inhibitor sensitivities. (A) IC50 of the mTORC1 inhibitor temsirolimus from COSMICs drug screening database in all cancer cell lines vs. ovarian cancer cell lines, horizontal lines indicate geometric mean. (B) MTT assay curves from AZD8055 (left) or everolimus (right) treatment on the aforementioned cell line panel. (C) Expression of p-AKT308, p-AKT473 and p-S6 after 24 (up) and 72h (down) exposure to 100 nM everolimus, AZD8055 or MLN0128 in the OCCC cell lines KOC7C or JHOC5 determined by Western blot. β-Actin was used as loading control. Results are representative from n=2 experiments. (D) Expression of p-AKT308, p-AKT473, p-S6 and (cleaved) PARP after 48h exposure to increasing concentrations of everolimus, AZD8055 and dactolisib in the OCCC cell lines KOC7C (up) and JHOC5 (down) determined by Western blot. β-Actin was used as loading control. Results are representative from n=2 experiments. (E) Everolimus, AZD8055, GDC0941 and selumetinib IC50 determined for 14 OCCC cell lines (ES2, KOC7C, SMOV2, JHOC5, RMG1, OVMANA, HAC2, OV207, OVTOKO, TOV21G, OVAS, OVCA429, TUOC1 and RMG2) and dactolisib IC50 determined for 7 OCCC cell lines (ES2, KOC7C, SMOV2, JHOC5, RMG1, OVMANA and HAC2) by MTT assay. Selumetinib IC50 of KOC7C was not reached at maximum used concentration of 25 uM. Horizontal lines indicate geometric mean. Data is derived from n{greater than or equal to}2 experiments. (F) Long-term proliferation assay after exposure to increasing concentrations of AZD8055 and dactolisib. Results are representative from n=3 experiments.

The genetic circuits that allow cancer cells to evade destruction by the host immune system remain poorly understood 1 – 3 . Here, to identify a phenotypically robust core set of genes and pathways that enable cancer cells to evade killing mediated by cytotoxic T lymphocytes (CTLs), we performed genome-wide CRISPR screens across a panel of genetically diverse mouse cancer cell lines that were cultured in the presence of CTLs. We identify a core set of 182 genes across these mouse cancer models, the individual perturbation of which increases either the sensitivity or the resistance of cancer cells to CTL-mediated toxicity. Systematic exploration of our dataset using genetic co-similarity reveals the hierarchical and coordinated manner in which genes and pathways act in cancer cells to orchestrate their evasion of CTLs, and shows that discrete functional modules that control the interferon response and tumour necrosis factor (TNF)-induced cytotoxicity are dominant sub-phenotypes. Our data establish a central role for genes that were previously identified as negative regulators of the type-II interferon response (for example, Ptpn2 , Socs1 and Adar1 ) in mediating CTL evasion, and show that the lipid-droplet-related gene Fitm2 is required for maintaining cell fitness after exposure to interferon-γ (IFNγ). In addition, we identify the autophagy pathway as a conserved mediator of the evasion of CTLs by cancer cells, and show that this pathway is required to resist cytotoxicity induced by the cytokines IFNγ and TNF. Through the mapping of cytokine- and CTL-based genetic interactions, together with in vivo CRISPR screens, we show how the pleiotropic effects of autophagy control cancer-cell-intrinsic evasion of killing by CTLs and we highlight the importance of these effects within the tumour microenvironment. Collectively, these data expand our knowledge of the genetic circuits that are involved in the evasion of the immune system by cancer cells, and highlight genetic interactions that contribute to phenotypes associated with escape from killing by CTLs.

Advanced stage ovarian clear cell carcinoma (OCCC) is poorly responsive to platinum-based chemotherapy and has an unfavorable prognosis. Previous studies revealed heterogeneous mutations in PI3K/AKT/mTOR and MAPK pathway nodules converging in mTORC1/2 activation. Here, we aimed to identify an effective low-dose combination of PI3K/AKT/mTOR pathway and MAPK pathway inhibitors simultaneously targeting key kinases in OCCC to preclude single-inhibitor initiated pathway rewiring and limit toxicity. Small molecule inhibitors of mTORC1/2, PI3K and MEK1/2 were combined at monotherapy IC20 doses in a panel of genetically diverse OCCC cell lines (n = 7) to determine an optimal low-dose combination. The IC20 dose triple combination reduced kinase activity in PI3K/AKT/mTOR and MAPK pathways, prevented single-inhibitor induced feedback mechanisms and inhibited short and long-term proliferation in all seven cell lines. Finally, this low-dose triple drug combination treatment significantly reduced tumor growth in two genetically characterized OCCC patient-derived xenograft (PDX) models without resulting in weight loss in these mice. The effectiveness and tolerability of this combined therapy in PDX models warrants clinical exploration of this treatment strategy for OCCC and might be applicable to other cancer types with a similar genetic background.

SWI/SNF chromatin remodeling complexes play an important role in the epigenetic regulation of chromatin structure and gene transcription. Mutual exclusive subunits in the SWI/SNF complex include the DNA targeting members ARID1A and ARID1B as well as the ATPases SMARCA2 and SMARCA4. SWI/SNF complexes are mutated across many cancer types. The highest mutation incidence is found in ARID1A, primarily consisting of deleterious mutations. Current advances have reported synthetic lethal interactions with the loss of ARID1A in several cancer types. In this review, we discuss targets that are only important for tumor growth in an ARID1A mutant context. We focus on synthetic lethal strategies with ARID1A loss in ovarian clear cell carcinoma, a cancer with the highest ARID1A mutation incidence (46-57%). ARID1A directed lethal strategies that can be exploited clinically include targeting of the DNA repair proteins PARP and ATR, and the epigenetic factors EZH2, HDAC2, HDAC6 and BRD2.

Current treatment for advanced stage ovarian clear cell cancer is severely hampered by a lack of effective systemic therapy options, leading to a poor outlook for these patients. Sequencing studies revealed that ARID1A is mutated in over 50% of ovarian clear cell carcinomas. To search for a rational approach to target ovarian clear cell cancers with ARID1A mutations, we performed kinome-centered lethality screens in a large panel of ovarian clear cell carcinoma cell lines. Using the largest OCCC cell line panel established to date, we show here that BRD2 inhibition is predominantly lethal in ARID1A mutated ovarian clear cell cancer cells. Importantly, small molecule inhibitors of the BET (bromodomain and extra terminal domain) family of proteins, to which BRD2 belongs, specifically inhibit proliferation of ARID1A mutated cell lines, both in vitro and in ovarian clear cell cancer xenografts and patient-derived xenograft models. BET inhibitors cause a reduction in the expression of multiple SWI/SNF members including ARID1B, providing a potential explanation for the observed lethal interaction with ARID1A loss. Our data indicate that BET inhibition may represent a novel treatment strategy for a subset of ARID1A mutated ovarian clear cell carcinomas.

Abstract Purpose: Advanced-stage ovarian clear cell carcinoma (OCCC) is unresponsive to conventional platinum-based chemotherapy. Frequent alterations in OCCC include deleterious mutations in the tumor suppressor ARID1A and activating mutations in the PI3K subunit PIK3CA. In this study, we aimed to identify currently unknown mutated kinases in patients with OCCC and test druggability of downstream affected pathways in OCCC models. Experimental Design: In a large set of patients with OCCC (n = 124), the human kinome (518 kinases) and additional cancer-related genes were sequenced, and copy-number alterations were determined. Genetically characterized OCCC cell lines (n = 17) and OCCC patient–derived xenografts (n = 3) were used for drug testing of ERBB tyrosine kinase inhibitors erlotinib and lapatinib, the PARP inhibitor olaparib, and the mTORC1/2 inhibitor AZD8055. Results: We identified several putative driver mutations in kinases at low frequency that were not previously annotated in OCCC. Combining mutations and copy-number alterations, 91% of all tumors are affected in the PI3K/AKT/mTOR pathway, the MAPK pathway, or the ERBB family of receptor tyrosine kinases, and 82% in the DNA repair pathway. Strong p-S6 staining in patients with OCCC suggests high mTORC1/2 activity. We consistently found that the majority of OCCC cell lines are especially sensitive to mTORC1/2 inhibition by AZD8055 and not toward drugs targeting ERBB family of receptor tyrosine kinases or DNA repair signaling. We subsequently demonstrated the efficacy of mTORC1/2 inhibition in all our unique OCCC patient–derived xenograft models. Conclusions: These results propose mTORC1/2 inhibition as an effective treatment strategy in OCCC. Clin Cancer Res; 24(16); 3928–40. ©2018 AACR.

Abstract Introduction: High stage ovarian clear cell carcinoma (OCCC) is less responsive to chemotherapy and has a worse prognosis than other ovarian cancer subtypes. The most frequently mutated genes in OCCC have been characterized in recent years. These include deleterious mutations in the tumor suppressor gene ARID1A, epigenetic silencing and deleterious mutations in the PI3K antagonist PTEN as well as activating mutations in the PI3K catalytic subunit PIK3CA. Yet, therapeutic strategies that utilize these genetic aberrations are lacking. In the present study, we aimed to identify and validate new kinase targets in OCCC. Methods: To determine new mutations and copy number gains and losses, kinome sequencing was performed on DNA isolated from tumor samples (n=124) and matched controls (n=47) and high-coverage SNP analysis was implemented for 109 of the collected OCCC tumor samples. Inhibitor sensitivity screening was performed on 17 OCCC cell lines to validate potential kinase targets, following in vivo validation in OCCC patient-derived xenografts (PDX). Results: Mutations in ARID1A, PIK3CA, PTEN, KRAS and TP53 corresponded to frequencies found in literature. Most identified mutations could be designated to the PI3K/Akt/mTOR and MAPK signal transduction pathway, and HER family of receptor tyrosine kinases. The PI3K pathway related genes PIK3R1, ERBB3 and AKT1 were significantly mutated in our dataset and not described in OCCC before. EGFR, ERBB2 and PIK3CA emerged as most frequently amplified kinases. Combining mutations and copy number alterations in these pathways, 91% of all tumors were affected. The highly positive staining of p-S6 (90% of OCCC tumors) indeed suggests high mTORC1/2 activity in these tumors. Inhibitor screening demonstrated subsets of cell lines to be sensitive to EGFR or ERBB2 inhibition. Furthermore, the vast majority of OCCC cell lines was susceptible to inhibition of PI3K and MAPK downstream target mTORC1/2 using AZD8055, which efficacy was further demonstrated in several OCCC PDX models. In contrast, mTORC1 inhibition alone was less effective. Conclusion: Most kinase mutations and copy number alterations in OCCC can be related to PI3K and MAPK pathway activation. In vitro and in vivo data suggest mTORC1/2 inhibition as a novel treatment strategy in OCCC. Combinations of EGFR or ERBB2 inhibition with mTORC1/2 targeted therapy can be envisioned for OCCC tumors with EGFR or ERBB2 alterations. Supported by a grant from the Dutch Cancer Foundation: RUG 2012-5477 Citation Format: J.j. Caumanns, G.b.a Wisman, K Berns, T. Tomar, R.s.n. Fehrmann, R Bernards, A.g.j. Van der Zee, S De Jong. Integrative kinome exome sequencing and copy number profiling of ovarian clear cell carcinoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 1219. doi:10.1158/1538-7445.AM2017-1219