Synthetic Lethal Screens Identify Vulnerabilities in GPCR Signaling and Cytoskeletal Organization in E-Cadherin–Deficient Cells

E-cadherin is a cell-to-cell adhesion protein that is localized at the adherens junction of all epithelial cells (1). Other than its roles in cell adhesion, E-cadherin is involved in establishing and maintaining cell polarity and differentiation, the organization of cell migration and architecture and the mediation of signaling through various proliferation and survival pathways, including WNT and EGFR (2, 3).

Abrogation of expression of the E-cadherin gene (CDH1) by mutation, deletion, or promoter hypermethylation is a feature common to many epithelial tumors and its downregulation is the hallmark of both diffuse gastric cancer (DGC) and lobular breast cancer (LBC; refs. 1, 4–6). Disrupting E-cadherin's expression or localization has a pronounced impact on a cell's cytoskeletal structure, with changes including misalignment of the microtubule and actin cytoskeletons, defects in cell migration and irregularities in the orientation of the mitotic spindle (7–9).

Germline CDH1 mutations are responsible for hereditary diffuse gastric cancer (HDGC), a cancer syndrome characterized by the highly penetrant, early onset of multifocal DGC and an elevated rate of LBC. In HDGC, CDH1 inactivation is an initiating event that may be related to abnormal mitotic spindle orientation resulting in daughter cells being displaced into the lamina propria, outside the epithelial plane (10–12). In other cancer types, its downregulation is considered to be a late event that promotes increased invasive capacity, frequently through association with the epithelial–mesenchymal transition (13).

Although E-cadherin is a tumor suppressor protein that is lost from the cancer cell and therefore not a conventional drug target, the downregulation of such a multifunctional protein during tumorigenesis would be predicted to create vulnerabilities in the cell which are targetable using a synthetic lethal approach. In the context of drug development, synthetic lethality can be defined as a drug that reduces cell viability or fitness only in cells carrying a specific mutation. The utility of synthetic lethal targeting of tumor suppressor genes is well illustrated clinically by olaparib, an inhibitor of the DNA repair enzyme PARP. Olaparib elicits strong clinical responses in breast and ovarian cancer patients who harbor inactivating mutations in the homologous recombination dsDNA repair genes BRCA1/2 (14, 15).

In addition to providing new therapeutic avenues for the treatment of sporadic epithelial cancers, synthetic lethal targeting of E-cadherin–deficient cells also has the potential to improve the clinical management of HDGC. To identify druggable synthetic lethal vulnerabilities in E-cadherin–deficient cells, we have conducted both a genome-wide siRNA synthetic lethal screen and a four thousand compound known drug screen on isogenic breast MCF10A cells with and without CDH1 expression. Together, these screens have identified multiple druggable targets that suggest new therapeutic strategies for the treatment of E-cadherin–deficient cancers and the chemoprevention of HDGC.

The MCF10A breast cell line and its paired isogenic MCF10A CDH1^−/− line (here designated CDH1^−/−) were obtained from Sigma Aldrich in 2011 (parental line ATCC CRL-10317) and had been authenticated using short terminal repeat analysis. CDH1^−/− had been created by homozygous deletion of 4 bp from exon 11 of the CDH1 gene. The lines were resuscitated within one week of receipt and early passage cells (passage 3–7) were aliquoted and frozen. All experiments were conducted with cells between passages 6–15 in DMEM F12 media with glutamate, 5% horse serum, 20 ng/mL EGF, 0.5 μg/mL hydrocortisone, 100 ng/mL cholera toxin, and 10 μg/mL insulin. Characterization of the cell line pair is described elsewhere (7).

Cells were transfected with siRNAs from the Dharmacon SMARTpool whole genome protein-coding siRNA library (RefSeq 27) housed in the Victorian Center for Functional Genomics. Each SMARTpool contained four siRNAs that targeted different regions of each gene in one well. Each reaction, in a white walled, clear-bottom 384-well plate format, contained 0.125% DharmaFECT 3 (0.05 μL), 27.4% OptiMEM (Invitrogen), and 40 nmol/L of the siRNA SMARTpool (total volume 37.5 μL). Cells were reverse transfected and seeded onto the siRNA cocktail at a density of 700 MCF10A cells per well and 900 CDH1^−/− cells/well to enable the two cell lines to reach confluence at the same time point (72 hours after seeding). The following controls were included in each plate: the death controls siEGFR and siPLK1 (4 wells each), a synthetic lethal control siCTNNB1 (6 wells) and two negative controls, siRISC free and mock (lipid only, 9 wells each). After 24 hours, the media was replaced and at 72 hours 10 μL CellTiter-Glo was added to each well (final concentration 1/5), shaken on an orbital shaker for 2 minutes and incubated for a further 30 minutes at room temperature before measuring luminescence using a Synergy H4 microplate reader (Biotek). siRNA was dispensed using a Caliper Sciclone ALH3000 (PerkinElmer). All other liquid handling steps were performed using a Biotek406 liquid handling workstation. Primary screen analysis was performed by normalizing genes in each cell line individually to the average mock value (across all screen plates) for the respective cell line. The level of increased kill was determined by the ratio of CDH1^−/− viability to MCF10A viability. Candidates with MCF10A viability ≥50% and a fold change ratio of ≤0.85 were considered synthetic lethal candidates. Selection of the final 500 genes chosen for secondary screening was based on further analysis of druggability and biologic relevance. Secondary siRNA screening was performed using four individual siRNAs targeting each gene arrayed in individual wells, separately using the same transfection conditions described above with a final siRNA concentration of 25 nmol/L. The secondary screen was analyzed using the same normalization strategy and cutoffs as the primary screen.

Dharmacon pGIPZ lentiviral shRNA mir30 plasmids were prepared from cultures using the Machery Nagel EasyPure Miniprep Kit. 293FT cells were cotransfected 24 hours after seeding with 18.6 μg pGIPZ, 9.6 μg PAX2, and 4.8 μg VSVG plasmids using 55.7 μL Lipofectamine 2000 (Invitrogen). Media was changed at 24 hours, and after a further 24 hours, viral particles were harvested by aspirating media, centrifugation at 3,000 rpm for 15 minutes to remove cellular debris and filtration through a 0.45 μmol/L polyvinylidene difluoride filter. Virus was aliquoted and snap frozen for subsequent use.

Viral titer was determined by seeding MCF10A cells at 4,000 cells per well and transducing with a 1/32 dilution of virus 24 hours later. Media was changed after 24 hours, and after a further 24 hours, GFP-expressing cells were quantitated over 5 fields at 10× magnification. The average number of transduced cells per well was used to determine the number of transducing units per mL.

For knockdown experiments, 1,000 MCF10A and 2,000 CDH1^−/− cells per well were seeded in black walled, clear bottomed 96-well plates and allowed to adhere overnight. The following day, virus was added at a multiplicity of infection (MOI) of 10. Media were changed at 24 hours and 1 μg/mL puromycin added. Seventy-four hours after transduction, media was aspirated and 1 μg/mL Hoechst 33342 and 0.5 μg/mL propidium iodide in PBS added. After 30 minutes, incubation plates were imaged with 4 fields per well on the Cytell (GE) at 4× magnification and 10 fields per well at 10× magnification using the “Cell Viability BioApp”. CellProfiler (16) was used to quantitate the total nuclei, as well as the proportion stained with propidium iodide.

RNA was extracted at 72 hours to determine gene knockdown using the RNAgem-PLUS Kit (ZyGem). cDNA was synthesized using the Primescript cDNA Synthesis Kit (Takara), and qPCR performed using Sybr Fast kit (KAPA) on an ABI7900HT with an enzyme activation step of 95°C for 3 minutes followed by 40 cycles of 95°C for 15 seconds, 57°C for 15 seconds, and 72°C for 15 seconds. GAPDH and PPIA were used as reference genes and results were analyzed using the efficiency method as described by Pfaffl (17).

Assay ready plates containing 20 nL of 4,057 compounds diluted in DMSO at four concentrations (2, 1, 0.5, and 0.25 μmol/L) were prepared by the Walter and Eliza Hall Institute (WEHI, Parkville, Victoria, Australia) High-Throughput Chemical Screening Facility. Cells were seeded directly onto these plates in a volume of 50 μL at a density of 1,000 cells per well for MCF10A and 1,200 cells per well for CDH1^−/− in 384-well clear bottom plates. After 48 hours, 20 μL of CellTiter-Glo (Promega) was added to each well, shaken for 2 minutes on an orbital shaker, and incubated for 30 minutes before reading. Values were normalised to the average DMSO control value of the whole screen for each cell line at each concentration.

The secondary drug screen was performed on 316 compounds selected from the primary screen. These were provided at 5 mmol/L and 11 1:1 dilutions were made with DMSO to create a 10 μmol/L to 10 nmol/L concentration range. Cells were seeded at a density of 700 cells per well for MCF10A and 900 cells per well for CDH1^−/− for 24 hours before 100 nL of each compound was added to the plates robotically. Thirty-two DMSO control wells were included in each plate. Plates were incubated for a further 48 hours before assaying with CellTiter-Glo.

To determine drug EC₅₀, cells were seeded in white walled, clear-bottom 384 well plates at a density of 800 cells per well in a volume of 45 μL. After 24 hours incubation, 5 μL of drug (rehydrated in DMSO to an 80 mmol/L stock and diluted in complete media) was added to each well. A DMSO control and a cell only control were used on each plate. The 11 drug concentrations used were dependent on the individual drug. After 48 hours of treatment, cell viability was assayed using either the CellTiter-Glo (Promega) or Alamar blue assay. All results were normalized to the average DMSO control for individual cell types. EC₅₀ values were calculated by plotting viability against the log drug concentration and fitting a nonlinear regression curve using Prism version 6.0 for Mac (GraphPad Software).

Drug time course assays were carried out by seeding equal numbers of MCF10A or CDH1^−/− at 4,000 cells per well in xCELLigence plates (Roche), except for crizotinib which was seeded at 2,000 cells per well. The xCELLigence system measures relative changes in electrical impedance in a well (“cell proliferation index”) which can be used as a measure of cell number. After 24 hours, three concentrations of each drug were added and growth followed in real time for a further 72 hours. Assays were carried out in duplicate and the data averaged. Time course assays were also performed on the IncuCyte Imaging System (Essen Bioscience). Cells were seeded at 4,000 cells per well in black walled clear-bottom 96-well plates. After 24 hours, five concentrations of each drug were added and 3 fields were imaged per well at 4× magnification every 2 hours for 48 hours. Plates were then removed and cell numbers determined using metabolic assays and nuclei imaging as described above.

For drug synergy studies, the combination index was determined using CompuSyn software (ComboSyn Inc) and the Chou–Talalay method (18). EC₅₀ was estimated for both drugs and cells treated with each drug alone and in combination at this concentration and concentrations 2- and 4-fold higher and lower. Viability was calculated by nuclei counting and normalized values imported into CompuSyn.

To identify genes potentially involved in a synthetic lethal interaction with E-cadherin, we conducted a genome-wide functional screen using the Dharmacon siGENOME SMARTpool library targeting 18,120 genes in a pooled format, with 4 individual siRNAs targeting each gene. Isogenic MCF10A cells with and without CDH1 expression were screened in parallel, and viability was assayed 72 hours after transduction using CellTiter-Glo. siPLK1 and siEGFR were used as positive death controls (Fig. 1A) and siCTNNB1 used as a positive synthetic lethal control, after showing a mild synthetic lethal effect in a pilot screen (Supplementary Fig. S1). Mock (lipid only) and RISC-free control (Dharmacon) were included as nontargeting controls. Little variability was observed between these controls; consequently, we normalized the values of each gene to the average screen-wide mock value within each cell line. Hits were selected on the basis of the ratio of cell viability between the MCF10A and CDH1^−/− cells 72 hours after transduction (Fig. 1B). We rejected siRNAs that were highly toxic to the MCF10A cells (a decrease in viability of ≥50%) and selected those for which the viability of CDH1^−/− cells decreased by ≥15% more than the corresponding knockdown in MCF10A cells. These targets were classed as CDH1 synthetic lethal candidate genes (List SL1, 2,437 genes; Supplementary Table S1). From this set, 501 genes were manually selected for secondary screening using criteria, including predicted druggability of the encoded proteins and biologic significance. The secondary screen was performed by deconvoluting the four individual siRNAs that consititute the SMARTpool. Using the same stringent threshold as the primary screen (i.e., MCF10A viability ≥50% and ≥15% more death in CDH1^−/−), 21 genes (5%) had 3 of 4 or 4 of 4 of the individual siRNAs show a synthetic lethal effect. One hundred eighty-three genes (44%) had 1 of 4 or 2 of 4 siRNAs validated. Fifty-one percent of genes were not validated (0/4) by this approach. Division of the selected genes into groups based on the strength of the synthetic lethal phenotype in the primary screen demonstrated that candidates which showed a greater viability differential between MCF10A and CDH1^−/− cells were more likely to be validated in the secondary screen (Fig. 1C and D). Of the candidates that reduced CDH1^−/− cell number by ≥25% more than MCF10A cells, 12% validated with 3 of 4 or 4 of 4 individual siRNAs, a validation rate comparable with other genome-wide RNAi studies (19, 20).

To search for functional enrichment in the 2,437 synthetic lethal candidates identified in the primary screen (List SL1), we conducted a gene ontology analysis using DAVID (21). Using the Functional Annotation Clustering tool, the most enriched functional cluster was a group of ten terms associated with G-protein–coupled receptor (GPCR) signaling (enrichment score = 10.01; Supplementary Table S2A). Accordingly, the two most significant biologic process terms (Supplementary Table S2B) were “G-protein–coupled receptor protein signaling pathway” (Benjamini-adjusted P value = 4.1 × 10⁻⁸) and “cell surface receptor linked signal transduction” (adj. P value = 1.7 × 10⁻⁵). These cell signaling processes were strongly reflected in the DAVID gene ontology molecular function terms (Supplementary Table S2C) which included peptide, neuropeptide, and purinergic nucleotide receptor activity and protein kinase activity (adj. P value 3.5 × 10⁻⁷ to 3.2 × 10⁻²).

Because our ultimate goal is to identify targeted drugs for E-cadherin–deficient tumors that have minimal toxicity against nonmalignant tissues, we also performed gene ontology analysis on a subset of genes from List SL1 whose corresponding siRNA SMARTpools had little or no impact on MCF10A cells (MCF10A viability ≥0.85 mock). 1,136 genes met this revised threshold (List SL2; Supplementary Table S3). Notably, this more stringent cutoff led to further enrichment of both the GPCR-associated functional cluster (enrichment score, 12.14) and the biologic process terms “GPCR protein signaling pathway” (adj. P value = 9.0 × 10⁻16) and “cell surface receptor linked signal transduction” (adj. P value = 4.8 × 10⁻¹⁰; Supplementary Table S2D).

Our earlier observation of abnormal cytoskeletal organization in E-cadherin–deficient MCF10A cells (7) prompted us to look for specific cytoskeletal functions associated with synthetic lethality. Although the adjusted P values for cytoskeletal-like terms in List SL1 and SL2 did not reach significance, DAVID functional clusters associated with each of cell motility, cell polarity, and cell adhesion were among the top five clusters observed in the group of synthetic lethal candidate genes which reduced MCF10A viability to <0.85 of the mock controls (enrichment scores, 2.18, 1.97, and 1.84, respectively; Supplementary Table S2E). This greater representation of cytoskeletal genes among synthetic lethal candidates that seriously affect MCF10A cell viability presumably reflects the essential role of many cytoskeletal proteins. Cyto- or nucleoskeletal functions that were represented in List SL1, often by multiple family members, included microtubule nucleation, organization, and function (e.g., TUBA1C, TUBG1, TUBB2A, MZT2A, ARPC3, NME4, NME7, MAST1, MAST2, MAST3, NEK1, NEK3, NEK4, NEK10, CLIP1, CLIP2, TEKT3, TEKT4, TEKT5), Rho-mediated motility (e.g., RHOB, RHOC, RHOH, RAC1, PAK2, TIAM1), linkages between the cytoskeleton and nucleoskeleton (e.g., SUN1, SUN2, SUN5, NSUN3, NSUN6, SYNE1), polarity (e.g., DLG1, DLG2, DLG4, DLG5, CELSR1, CELSR3), and actin filament organization and remodeling (e.g., AVIL, ARF6, CYTH2, CTYH3, CYTH4).

To validate the apparent synthetic lethal interaction between E-cadherin and the microtubule cytoskeleton, we selected three microtubule-associated genes, MAST2, MAP1B, and MAPRE3 for confirmation using lentiviral shRNA knockdown. MAST2 is a microtubule-associated serine–threonine kinase (22); MAP1B is a microtubule-binding and -stabilizing protein that can also interact with actin microfilaments (23) and MAPRE3 is a microtubule plus end-binding protein involved in regulating the dynamics of microtubules and their interactions with intracellular structures such as the cell cortex or mitotic kinetochore (24). These three genes had previously been validated in the secondary screen, with 3 of 4, 2 of 4, and 1 of 4 siRNAs, respectively, decreasing the viability of CDH1^−/− cells by at least 15% more than the MCF10A (Fig. 2A). shRNAs targeting these three genes resulted in 51%–86% mRNA knockdown in both cell lines (Fig. 2B). Seventy-two hours after transduction, cell viability was measured using nuclei counting (Fig. 2C). Knockdown of each of MAP1B, MAST2, and MAPRE3 resulted in 15%–29% more cell death in the CDH1^−/− cells compared with the MCF10A cells, confirming the siRNA data.

To determine whether the synthetic lethal phenotype observed with downregulation of microtubule-associated genes could be recapitulated using known inhibitors of microtubule function, we treated the isogenic MCF10A cell line pair with the microtubule stabilizing drug taxol and inhibitors of the microtubule-associated proteins aurora kinase A and aurora kinase B. Increasing concentrations of taxol (1–16 nmol/L) led to a small increase in cell death in CDH1^−/− cells compared with MCF10A cells (Supplementary Fig. S2). The aurora kinase A inhibitor alisertib and the aurora kinase B inhibitor danusertib both showed a minor synthetic lethal effect (Fig. 2D and E), with 18% (at 25 nmol/L) and 12% (50 nmol/L) more death in the CDH1^−/− line, respectively. Together, the RNAi and aurora kinase inhibitor data demonstrate that E-cadherin–deficient MCF10A cells are more vulnerable to disruption of specific microtubule-related functions than E-cadherin–expressing MCF10A cells.

Gene ontology analysis (21) identified >200 genes from our list of synthetic lethal candidates (List SL1) that were associated with GPCR protein signaling pathways. These candidates included several proteins involved in signal transduction from activated GPCRs such as the G protein subunits GNAS, GNAT1, GNG2, and GNG5, the membrane bound adenyl cyclase ADCY7 and the downstream signaling protein JAK2. ADCY7 validated in the secondary screen with 4 of 4 siRNAs recapitulating the SMARTpool phenotype (Fig. 3A). The synthetic lethal effect of the JAK2 siRNA SMARTpool was confirmed using lentiviral-mediated shRNA transduction. Both the siRNA SMARTpool and lentiviral shRNA knocked down JAK2 mRNA by >44% (Fig. 3B). Normalized cell numbers after transduction were significantly lower in the CDH1^−/− cells compared with the MCF10A cells for both the siRNA pool and the shRNA (Fig. 3C). To determine whether the synthetic lethality of JAK2 downregulation could be mimicked using a JAK2 antagonist, we treated the isogenic CDH1 MCF10A cell line pair with the JAK inhibitor LY2784544. Using the xCELLigence system to monitor cell growth in real time, LY2784544 had only a modest effect on MCF10A cells at 0.32, 0.63, and 1.25 μmol/L concentrations; however, the cell proliferation index was reduced in the CDH1^−/− cells in a concentration-dependent manner. Forty-eight hours after drug addition, the three concentrations of LY2784544 resulted in a 15%, 29%, and 51% reduction in CDH1^−/− cell number, respectively (Fig. 3D). In contrast, the three drug concentrations reduced MCF10A cell number by 0%, 11%, and 10%, respectively. A second JAK2 inhibitor, AG490, showed a similar synthetic lethal response in xCELLigence assays (data not shown). Further real-time assays using the IncuCyte replicated this effect (Supplementary Fig. S3). Nuclei counting confirmed a significant synthetic lethal effect at LY2784544 concentrations of 1.25 μmol/L (P = 0.005) and 0.63 μmol/L (P = 0.03) (Fig. 3E). LY2784544 resulted in cells becoming more spindle shaped, with an increased number of extended filopodia; this effect was particularly marked in the CDH1^−/− cells (Fig. 3F).

The presence of G protein subunits in list SL1 prompted us to examine CDH1 synthetic lethality using the Gαi and Gαo subunit inhibitor Pertussis toxin (25). Treatment of MCF10A and CDH1^−/− cells with 10, 100, and 200 ng/mL of Pertussis toxin over a period of 48 hours resulted in growth inhibition of both MCF10A and CDH1^−/− cell lines in a concentration-dependent manner (Fig. 3G). However, the MCF10A cells recovered and reached the same confluence as the PBS control after 48 hours, whereas the CDH1^−/− cells showed 20%–30% less confluence at that time point for the three drug concentrations.

To explore how E-cadherin loss alters sensitivity of MCF10A cells to other known drugs, we screened 4,057 compounds against the CDH1 MCF10A isogenic cell line pair. The compounds comprised the WEHI known drug library (3,600 compounds from the Tocriscreen Total library, the Prestwick Chemical Library and the “Lopac 1280” library), the Selleck Chemistry inhibitor library (326 compounds consisting of approximately half known drugs and half kinase inhibitors) and a kinase inhibitor library (131 compounds supplied by SYNthesis Medicinal Chemistry). The initial screen covered four drug concentrations ranging from 0.25 to 2 μmol/L, with cell viability measured at 48 hours after drug addition using the CellTiter-Glo assay. Potential synthetic lethal drugs were selected for further characterization if they met two criteria: (i) modest toxicity to MCF10A cells (a decrease in viability of no more than 30%) and (ii) a minimum of 15% greater reduction in CDH1^−/− viability compared with the MCF10A cells at one or more concentrations. Three-hundred sixteen compounds were selected for secondary analysis using an 11 point serial dilution from 10 μmol/L to 10 nmol/L. Twenty-one of 316 compounds in this secondary screen had EC₅₀ values that were 10%–50% lower in the CDH1^−/− cells compared with the MCF10A cells (Table 1). These included multiple histone deacetylase (HDAC) and PI3K inhibitors, crizotinib (an inhibitor of receptor tyrosine kinases c-MET, ALK, and ROS1), CGP 71683 hydrochloride (an inhibitor of the neuropeptide receptor NPY5R) and the guanine nucleotide exchange factor inhibitor Brefeldin A. The synthetic lethality of the majority of drug classes shown in Table 1 was supported by the siRNA primary screen data with one or more targets (or associated proteins) for each being included in List SL1 (Supplementary Table S4).

The E-cadherin synthetic lethal effects of crizotinib and several HDAC inhibitors were further characterized in time course and direct cell counting studies. Saracatinib, a c-SRC kinase inhibitor not included in the 11-point screen was also further examined because of a borderline effect in the original four-point screen. Crizotinib had little effect on the growth of MCF10A cells at 0.63, 1.25, and 2.50 μmol/L up to 48 hours after drug addition, as observed using the xCELLigence system. The same concentrations, however, reduced the growth of CDH1^−/− cells to 86%, 76%, and 46% of mock (Fig. 4A). Nuclei counting at 48 hours confirmed the synthetic lethal effect, although an inhibitory effect was observed on the MCF10A cells at all three concentrations (Fig. 4B). The difference between the nuclei counting method and the IncuCyte and xCELLigence methods is primarily due to cell density differences at full confluency that can only be determined by direct nuclei counting.

Treatment with saracatinib caused greater growth inhibition in the CDH1^−/− cells compared with the MCF10A cells at three different concentrations in two different assay systems, cell confluence (IncuCyte) and direct nuclei counting. In the IncuCyte system, a dose-dependent inhibition was observed in both isogenic cells with the CDH1^−/− cells demonstrating greater susceptibility (Fig. 4C). At 0.63 μmol/L, saracatinib had negligible effect on the confluence of MCF10A cells but caused a 26% inhibition of CDH1^−/− cells (relative to DMSO) after 48 hours. Similarly, differentials of 0.17 (P = 0.06), 0.30 (P = 0.02), and 0.15 (P = 0.10) were observed in normalized cell counts of the CDH1^−/− cells compared with the MCF10A cells at saracatinib concentrations of 0.63, 1.25, and 2.5 μmol/L, respectively (Fig. 4D).

Entinostat selectively targets class I HDACs, in particular HDAC1–3 (26). The 0.63, 1.25, and 2.5 μmol/L entinostat had a negligible effect on cell proliferation of MCF10A as determined using the xCELLigence system. In contrast, CDH1^−/− cells showed 18%, 67%, and 78% growth inhibition at these concentrations 48 hours after drug addition (Fig. 4E). Comparable results were obtained using the IncuCyte (Supplementary Fig. S3). Nuclei counting also showed a significant synthetic lethal effect across the three entinostat concentrations, although, as observed previously for crizotinib, reduced nuclei count was observed in both cell lines with increasing drug concentration using this more direct method. At concentrations of 0.63, 1.25, and 2.5 μmol/L, a cell viability differential of 0.13 (P = 0.04), 0.23 (P = 0.004), and 0.14 (P = 0.02) was observed between the CDH1^−/− cells and the MCF10A cells (Fig. 4F).

Vorinostat (SAHA) is a pan-HDAC inhibitor, acting on both class I and class II HDACs (26). Assays using the IncuCyte system showed preferential inhibition of CDH1^−/− cells over 48 hours of drug treatment at 0.63, 1.25, and 2.5 μmol/L, with little effect on the MCF10A cells (Fig. 4G). This effect was confirmed on the xCELLigence system (Supplementary Fig. S3). Similar to entinostat, direct nuclei counting showed a greater effect of vorinostat on the MCF10A cells than was observed using the real-time proliferation assay platforms. A significant cell viability differential was still observed between the CDH1^−/− and MCF10A cells with cell viability differentials of 0.21 (P = 0.02), 0.29 (P = 0.01), and 0.17 (P = 0.10) between CDH1^−/− and MCF10A at concentrations of 0.63, 1.25, and 2.5 μmol/L (Fig. 4H). The HDAC inhibitors mocetinostat and pracinostat showed comparable synthetic lethal effects when assayed at 0.63, 1.25, and 2.5 μmol/L on the xCELLigence system (Supplementary Fig. S4A and S4B). The class I/II HDAC inhibitor valproic acid also showed a minor synthetic lethal effect when assayed by direct nuclei counting (Supplementary Fig. S4C).

Previous reports of synergy between HDAC inhibitors and taxol (27–30) prompted us to test combinations of taxol and each of vorinostat and entinostat in our isogenic cell line pair. CompuSyn (ComboSyn Inc) was used to calculate EC₅₀ concentrations, and to determine the Combination Index. In contrast to other studies, we found no evidence for synergy between taxol and these HDAC inhibitors (Supplementary Fig. S2).

Synthetic lethality provides a potential method to target cancers carrying inactivating mutations in tumor suppressor genes such as the E-cadherin gene, CDH1. To provide an initial survey of E-cadherin's synthetic lethal interactions, we conducted a genome-wide functional screen of nonmalignant, isogenic MCF10A cells with and without E-cadherin expression. Thirteen percent of the 18,120 genes in the siRNA screen met our threshold for synthetic lethality of at least 15% more death in the CDH1^−/− cells than the wild-type MCF10A cells. Although this threshold is low stringency and subsequently not highly specific, it is clear that E-cadherin deficiency creates large numbers of vulnerabilities in nonmalignant cells which are exposed by genetic knockdown of additional genes.

Gene ontology analysis identified a striking enrichment for GPCR signaling proteins among the synthetic lethal candidates. Notably, this enrichment was greater for hits that showed minimal impact on the viability of the E-cadherin–expressing MCF10A cells, suggesting that drug targeting of GPCR signaling in CDH1-mutant tumors may be a means to obtain clinical gain while minimizing collateral damage to normal tissues. The nature of the synthetic lethal relationship between E-cadherin and GPCR signaling is not yet known, although the functional diversity of the candidate synthetic lethal GPCR signaling proteins would suggest that the interaction involves a common downstream mechanism such as interplay with the actin and microtubule cytoskeletons (31–33). The importance of cytoskeletal functions to the E-cadherin synthetic lethal phenotype is supported by the abundance of cytoskeletal genes associated with synthetic lethality in our primary siRNA screen. These genes were involved in all aspects of cytoskeletal function, including the nucleation, organization, and function of microtubules mitotic spindle organization and control, linkages between the cytoskeleton and nucleoskeleton, polarity, actin filament organization, vesicle transport, focal adhesion kinase signaling, and Rho-mediated motility (8, 34–36). The increased vulnerability of E-cadherin–deficient MCF10A cells to knockdown of so many diverse but inter-related functions is consistent with the widespread disorganization of cytoskeletal networks observed in the CDH1^−/− cells (7). Notably, after CDH1 and TP53, RHOA is the most commonly mutated gene in DGC (37–39), emphasizing the importance of dysregulated cytoskeletal function to development of the diffuse phenotype.

The increased vulnerability of E-cadherin–deficient MCF10A cells to RNAi knockdown of cytoskeletal and GPCR signaling genes was supported by the increased sensitivity of these cells to antagonists of multiple protein families associated with GPCR signaling and cytoskeletal function, including HDACs (40, 41), JAK (42), aurora kinases (43), c-SRC tyrosine kinase (44), G-protein subunits and PI3K (45). The enrichment for GPCR signaling genes among our synthetic lethal candidates does not, however, exclude the possibility that other signaling functions, such as the cytokine responses of JAK2, may also be associated with synthetic lethality. EC₅₀ differences between MCF10A and CDH1^−/− cells for these known drugs were on average only approximately 25% less in the CDH1^−/− cells. This small differential is to be expected as neither MCF10A or CDH1^−/− cells are tumorigenic and therefore these lines cannot be distinguished by the presence/absence of addiction to the targeted pathways. Synthetic lethal drugs which are significantly more potent in E-cadherin–deficient cells will be more readily identified using fit-for-purpose, high-throughput compound screens across the MCF10A and other CDH1 isogenic cell line pairs. Differences between the E-cadherin–expressing and E-cadherin–deficient cells may also be more pronounced in phenotypes other than cell viability, such as invasive capability.

This research demonstrates for the first time that loss of the tumor suppressor protein E-cadherin creates druggable vulnerabilities in cells. It remains to be determined whether any of the observed drug sensitivities in CDH1^−/− cells will be robust to the genetic dysregulation of advanced tumors, and therefore able to provide additional clinical benefit in the treatment of sporadic CDH1-mutant tumors. Instead, the observed sensitivities may have more near term application to HDGC chemoprevention. The natural history of cancer development in CDH1 germline mutation carriers involves the development of multifocal lobular carcinoma in situ (46, 47) and tens to hundreds of gastric stage T1a signet ring cell carcinomas before the onset of advanced disease (11, 48, 49). The high multiplicity of these early stage foci argues against additional genetic hits being required for their initiation. These early breast and gastric cancers are therefore relatively genetically homogenous and distinguished from normal tissue predominantly by the cellular changes associated with deficiency of E-cadherin. As a consequence, the E-cadherin synthetic lethal interactions identified in the nonmalignant breast MCF10A cells provide strong leads for drugs that may eliminate early-stage disease in germline CDH1 mutation carriers, potentially providing a new clinical management option for HDGC families.

No potential conflicts of interest were disclosed.

Conception and design: A. Chen, K.J. Simpson, P. Guilford

Development of methodology: B.J. Telford, A. Chen, H. Beetham, J. Frick, T.P. Brew, A. Single, T. Godwin, K.J. Simpson, P. Guilford

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B.J. Telford, A. Chen, J. Frick, T.P. Brew, A. Single, T. Godwin, K.J. Simpson, P. Guilford

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.J. Telford, A. Chen, J. Frick, T.P. Brew, C.M. Gould, A. Single, K.J. Simpson, P. Guilford

Writing, review, and/or revision of the manuscript: B.J. Telford, A. Chen, H. Beetham, T.P. Brew, A. Single, K.J. Simpson, P. Guilford

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B.J. Telford, A. Chen, H. Beetham, C.M. Gould, P. Guilford

Study supervision: A. Chen, K.J. Simpson, P. Guilford

VCFG technical expertise was provided by Daniel Thomas and Yanny Handoko. Lentiviral expertise and training was provided by Dr. Stephanie Hughes and Hollie Wicky. Compound screening was performed at the Walter Eliza Hall Institute under the supervision of Dr. Kurt Lackovic.

This work was supported by the New Zealand Health Research Council (grant number 11/513 to P. Guilford) and the University of Otago (PhD Fellowships to B. Telford, H. Beetham, and A. Single). Additional funding support was received from No Stomach for Cancer Inc, the Degregario Family Foundation and the New Zealand Breast Cancer Research Partnership. The Victorian Centre for Functional Genomics (to K. Simpson) is funded by the Australian Cancer Research Foundation (ACRF), the Victorian Department of Industry, Innovation and Regional Development (DIIRD), the Australian Phenomics Network (APN) supported by funding from the Australian Government's Education Investment Fund through the Super Science Initiative, the Australasian Genomics Technologies Association (AMATA), the Brockhoff Foundation, and the Peter MacCallum Cancer Centre Foundation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.