Loss of ARID1A expression sensitizes cancer cells to PI3K- and AKT-inhibition.

ARID1A mutations are observed in various tumors, including ovarian clear cell (OCCC) and endometrioid carcinomas, endometrial, and breast carcinomas. They commonly result in loss of ARID1A-protein expression and frequently co-occur with PI3K/AKT-pathway activating mechanisms. The aim of this study was to test the hypothesis as to whether PI3K/AKT-pathway activation is a critical mechanism in ARID1A-mutated tumors and if consequently ARID1A-deficient tumors show increased sensitivity to treatment with PI3K- and AKT-inhibitors. Upon ARID1A knockdown, MCF7 breast cancer cells and primary MRC5 cells exhibited a significantly increased sensitivity towards the AKT-inhibitors MK-2206 and perifosine, as well as the PI3K-inhibitor buparlisib. Knockdown of ARID1A in MCF7 led to an increase of pAKT-Ser473. AKT-inhibition with MK-2206 led to increased apoptosis and to a decrease of pS6K in ARID1A-depleted MCF7 cells but not in the controls. In five OCCC cell lines ARID1A-deficiency correlated with increased pAKT-Ser473 levels and with sensitivity towards treatment with the AKT-inhibitor MK-2206. In conclusion, ARID1A-deficient cancer cells demonstrate an increased sensitivity to treatment with small molecule inhibitors of the PI3K/AKT-pathway. These findings suggest a specific requirement of the PI3K/AKT pathway in ARID1A-deficient tumors and reveal a synthetic lethal interaction between loss of ARID1A expression and inhibition of the PI3K/AKT pathway.


INTRODUCTION
Mutations in the gene encoding the AT-rich interacting domain containing protein 1A (ARID1A) are frequently observed in a wide variety of gynecological and non-gynecological cancers [1,2]. These occur in approximately 50% of endometriosis-associated ovarian clear cell (OCCC) and 30% of endometrioid ovarian carcinomas (EnOC) [3,4], in endometrial carcinomas, with a loss of expression in 20-30% depending on the histological subtype [5,6], as well as in breast carcinomas (mutations in 4-35%) [7,8]. Non-gynecological carcinomas with frequent ARID1A mutations include pancreatic carcinomas (mutations in 8-45%) [9,10], gastric adenocarcinomas (mutations in 8-29%) [11][12][13], hepatocellular carcinomas (mutations in 10-17%) [14][15][16], as well as clear cell renal cell carcinomas [17,18]. The majority of the mutations lead to a loss of the ARID1A encoded protein [3], also referred to as BAF250a or p270, which is a subunit of the SWI/SNF chromatin remodeling complex [2]. Although ARID1A has recently been identified as a bona fide tumor suppressor gene and is currently being intensively investigated, the knowledge about the function and the consequences of a loss of expression of this protein is relatively limited [2].
Interestingly, ARID1A mutations frequently coexist with activating mutations of PIK3CA [12,19] and/or loss of PTEN expression [20], which both lead to a downstream activation of the PI3K/AKT pathway. Furthermore, it has recently been shown in endometrial cancer that loss of ARID1A expression leads to an increased phosphorylation of AKT at Ser-473 [21].
Similarly, increased AKT phosphorylation has also been reported in OCCC tissue samples with loss of ARID1A expression when concomitant PIK3CA mutations and loss of PTEN expression were excluded [22]. These observations strongly suggest interdependency between ARID1A mutations and PI3K/ AKT pathway activation, indicating that tumor cells with loss of ARID1A expression may be dependent on constitutive activation of the PI3K/AKT-pathway and consequently may also be more vulnerable to its inhibition [23]. This is of considerable clinical relevance since loss of ARID1A expression may be predictive for a favorable treatment response to small molecule inhibitors of the PI3K/AKT-pathway, which are currently under clinical investigation.
In this study, we demonstrate that depletion of ARID1A protein expression significantly increases the sensitivity of cancer cells towards PI3K-and AKTinhibitors, which is reflected by increased rates of apoptosis in treated ARID1A-depleted cells. Our findings suggest a dependency of ARID1A-mutated tumors on activating mechanisms of the PI3K/AKT-pathway, which may be exploited therapeutically.

ARID1A-deficient cell lines demonstrate increased sensitivity to treatment with the AKT-inhibitors MK-2206 and perifosine
ARID1A-depleted MCF7 cells were found to be significantly more sensitive to treatment with the AKTinhibitors MK-2206 and perifosine, as demonstrated in Figure 1a and 1b. The LogIC50 values for treatment with MK-2206 were -7.411 ±0.211 in ARID1A-depleted MCF7 cells, when compared to -4.841 ±0.273 in the controls (p<0.0001). This represents a more than 370fold increased AKT inhibitor sensitivity of ARID1Adepleted MCF7 cells as compared to the wild-type control. LogIC50 for perifosine in ARID1A-depleted MCF7 cells was -4.984 ±0.133, compared to the LogIC50 of -3.774 ±0.423 (p=0.0001) for the controls, which means an approximately 16-fold increased sensitivity.
We performed the same experiment in the primary fibroblast cell line MRC5 to verify our observations in a cell line without any oncogenic mutations and, similarly, we found an increased sensitivity of ARID1A-depleted MRC5 cells towards treatment with MK-2206 with LogIC50 of -6.307 ±0.173 as compared to -5.316 ±0.236 for the control (p=0.002). This meant an approximately 10fold increase in sensitivity. For treatment with perifosine the LogIC50 values were -5.867 ±0.135 in ARID1Adepleted MRC5 cells, as compared to -5.024 ±0.199 in the controls: equivalent to an increase in sensitivity of approximately 7-fold in ARID1A-depleted MRC5 cells (p=0.004).
As a control, no significant difference in sensitivity to treatment with MK-2206 or perifosine was found in the ARID1A-deficient OCCC cell line OVSAYO after transfection with ARID1A siRNA.

ARID1A-deficient cell lines demonstrate increased sensitivity to treatment with the PI3K-inhibitor buparlisib
The sensitivity to treatment with the PI3K inhibitor buparlisib in ARID1A-depleted MCF7 and MRC5 cells was shown to increase significantly with comparison to the controls, but less pronounced than for inhibition of AKT ( Figure 1c). The LogIC50 values were -6.780 ±0.150 for ARID1A-depleted MCF7 cells, as compared to -6.252 ±0.184 for the controls (approximately a 3-fold increase; p=0.025) and -6.712 ±0.154 for ARID1A-depleted MRC5 as compared to -5.950 ±0.128 for the controls (p=0.002), representing an almost 6-fold increase in sensitivity. The ARID1A-deficient control cell line, OVSAYO, did not show an increased sensitivity for buparlisib upon transfection with ARID1A siRNA.

Loss of ARID1A expression increases AKT phosphorylation at Ser-473 and downstream phosphorylation of p70 S6 kinase in MCF7 and MRC5 cells
First, we verified that loss of ARID1A increases phosphorylation of AKT at Ser-473 (pAKT-Ser 473 ) as has been described in endometrial cancer cell lines [21]. In both MCF7 and MRC5 cell lines, ARID1A knockdown by siRNA led to an increased phosphorylation of pAKT-Ser 473 as well as an increased downstream phosphorylation of p70S6kinase (pS6K) as shown in Figure 2a and 2b.

AKT-inhibition leads to apoptosis in ARID1Adeficient cells
We next investigated if inhibition of AKT leads to increased apoptosis in ARID1A-deficient cells. No cleavage of PARP1 (a marker of apoptosis) was observed in the MCF7 control cells (transfected with siLUC) with or without treatment with the AKT-inhibitor, MK-2206. In contrast, PARP1 cleavage was discretely detectable in ARID1A-depleted MCF7 cells and substantially increased after treatment with the AKT-inhibitor MK-2206 ( Figure  2a). Similar observations were made for MRC5 ( Figure  2b). Consistently, using a TUNEL-assay technique we found that apoptosis was markedly increased in ARID1Adepleted MCF7 and MRC5 cells treated with MK-2206, in comparison to the control cells that expressed normal levels of ARID1A ( Figure 3). Upon siARID1A transfection and treatment with MK-2206, PARP1 cleavage was not increased in the ARID1A-deficient control cell line, OVSAYO, which confirmed that the increased apoptosis rate in response to the AKT-inhibitor treatment was specifically due to the knockdown of the ARID1A gene and not to other unspecific factors that would be indirectly related to the transfection method ( Figure 2c).

Knockdown of AKT abrogates the increased proliferation rate of ARID1A-depleted MCF7 cells
ARID1A depletion led to an increased proliferation of MCF7 cells in comparison to the controls (Figure 2d). Knockdown of only AKT1 reduced measurable pAKT-Ser 473 -and AKT-levels, and led to a decreased level of pS6K, but did not lead to a difference in the amount of viable MCF7 cells after 5 days (Figure 2d, 2e). In contrast, combined knockdown of ARID1A and AKT1 completely abrogated the increased proliferation in ARID1A-depleted MCF7 cells (Figure 2d).

Sensitivity to treatment with MK-2206 in OCCC
Loss of ARID1A expression correlated with increased pAKT-Ser 473 in five OCCC cell lines ( Figure  4b, 4c). The cell lines, OVSAYO, OVISE, and HCH-1, did not express detectable levels of ARID1A and showed high sensitivity towards treatment with the AKT-inhibitor, MK-2206, whereas the two OCCC cell lines with intact ARID1A expression were resistant towards the same

DISCUSSION
The results of this study demonstrate an interdependency of ARID1A and the PI3K/AKT pathway, which results in significantly increased sensitivity of ARID1A-deficient cancer cells to PI3Kand AKT-inhibition. We thus demonstrate for the first time a principle of synthetic lethality between ARID1A deficiency and inhibition of the PI3K/AKT pathway in two cellular model systems. MCF7 is a well-established breast carcinoma cell line and MRC5 a commonly used human primary lung fibroblast cell line. Although MCF7 contains a moderately activating mutation of PIK3CA [24,25], depletion of ARID1A by siRNA considerably increased phosphorylation of AKT at Ser-473 in this cell line, as described previously in endometrial carcinoma cell lines [21]. The normal primary fibroblast cell line, MRC5, that does not contain any known mutations [26] confirmed our results, thus minimizing the likelihood of secondary effects by other mutations in the ARID1A knockdown experiments.
Mutations in the ARID1A gene are a frequent event occurring in a wide variety of gynecological and nongynecological cancers [23]. They usually lead to a loss of expression of the ARID1A encoded protein [3], and are thus not directly targetable [2]. Indirect therapeutic approaches using the principle of synthetic lethality in regard to members of the SWI/SNF chromatin remodeling complexes are an appropriate alternative strategy [27,28]. Thus, a recent study demonstrated a synthetic lethal relationship between inactivating mutations of the homologs ARID1A and ARID1B, both in primary and cancer cells. ARID1B depletion in the context of loss of ARID1A expression led to a significant decrease in cell proliferation revealing this homologue as a relative vulnerability in ARID1A-mutant cell lines. Despite the substantial relevance of these observations for potential future approaches, the immediate clinical relevance is limited due to the fact that there are currently no substances available that would therapeutically inhibit ARID1B expression [28].
An interdependency between ARID1A mutations and PIK3CA is strongly supported by the supplementary data of the study of Helming et al., since PIK3CA (just after ARID1B) scored as the second top candidate preferentially necessary for the growth of ARID1A-mutated cancer cell lines in the project, 'Achilles' [28]. A dependency of ARID1A-mutated tumors on activation of the PI3K/AKT pathway is further supported by a study in ARID1Aknockout and ARID1A/PTEN-double-knockout mice, where it was observed that only mice with simultaneous loss of ARID1A and PTEN expression developed poorly differentiated ovarian tumors, in contrast to mice with loss of only ARID1A that did not develop ovarian tumors [29]. This indicates that ARID1A-mutated cell clones require a second-hit mutation in order to transform into cancer, which may explain why loss of ARID1A expression was observable in a subset of non-atypical benign endometriosis in one of our previous studies [30].
Interestingly, ARID1A-mutated OCCC have been reported to be associated with resistance to conventional platinum-based chemotherapy regimens in a recent study [31]. If this observation is confirmed in larger collectives, an analysis of the ARID1A expression status, which is easily feasible by immunohistochemistry, will possibly become an important predictive tool to assess the probability of resistance to conventional platinum based chemotherapy and, as we suggest in this study, increased sensitivity to PI3K-and/or AKT-inhibitors. PI3K/AKTinhibitors, which are a promising tool to overcome chemotherapy resistance, may be selectively added to the treatment strategy in patients with ARID1A-deficient tumors.
In conclusion, ARID1A deficiency leads to a significantly increased sensitivity towards PI3K-and AKT-inhibition in tumor cells in vitro. The findings suggest a specific requirement of the PI3K/AKT pathway in ARID1A-deficient cancer cells and reveal a synthetic lethal interaction between loss of ARID1A expression and inhibition of the PI3K/AKT pathway.

Transient siRNA transfections
Transient knockdown of ARID1A was performed with four different ARID1A siRNAs (J-017263-05, -06, -07, and -08, ON-TARGETplus Human ARID1A siRNA, Dharmacon, Thermo Scientific, Thermo Fisher Scientific Inc., Waltham, MA, USA) using RNAimax (Gibco) according to the manufacturer's instructions. The control cells were transfected with siRNA against luciferase (sense): CGUACGCGGAAUACUUCGATTdTd (Microsynth AG, Baldach, Switzerland). The AKT1 knockdown control experiments were performed with a siRNA smart-pool (L-003000-00, ON-TARGET plus Human AKT1 (207) siRNA, Dharmacon). Cells were transfected overnight and the medium was replaced the following morning. The size of culture dishes and length of transfection is indicated for the specific experiments .

Treatment with AKT-and PI3K-inhibitors
The AKT-inhibitors MK-2206, perifosine, and the pan-PI3K-inhibitor buparlisib (BKM120) were all obtained from Selleckchem.com (Selleck Chemicals, Houston, TX, USA). The inhibitors were diluted according to the manufacturer's protocol (MK-2206 and buparlisib in DMSO and perifosine in sterile ddH 2 O) and prepared freshly for each treatment. Vehicle substance was used in the highest corresponding concentration for the treatment of the controls.

Drug sensitivity assays
Drug sensitivity assays were performed according to the drug manufacturer's recommendations (Selleckchem. com). In brief, cells were seeded in 96-well plates at concentrations of 1000-2000 cells per well (MRC5 cells were seeded at a density of 10'000 cells per well). On the following day, cells were transiently transfected overnight with the corresponding siRNAs and then treated with the corresponding drugs at the indicated concentrations or vehicle substance as described above. Treatment was repeated after 2-3 days. After a total period of 4-5 days' treatment, the medium was replaced by phenol-red free medium and the quantity of viable cells was assessed using a MTS assay (Cell Titer Aqueous, Promega, Madison, WI, USA) according to the manufacturer's protocol. Plate absorbances were obtained using a spectrophotometer (Epoch Microplate Spectrophotometer, BioTek, Winooski, VT, USA). Each experiment was performed on at least two separate occasions for every unique ARID1A siRNA sequence and each of them run in triplicates.

TUNEL assay and fluorescence microscopy
A TUNEL assay (In Situ Cell Death Detection Kit, Fluorescein, Roche, Basel, Switzerland) based on labelling of DNA strand breaks was used to directly visualize apoptosis in the treated cell cultures. In brief, cells were directly plated on microscopy coverslips transfected with siRNAs and treated for 60h with MK-2206 or vehicle substance as described above. Positive controls were obtained by treatment of fixated non-apoptotic cells with 30U/l recombinant DNase (in 1% (w/v) BSA / PBS) for 1h immediately prior to the addition of the incubation mix. Negative controls were obtained by omission of the TUNEL enzyme in the incubation mix. The coverslips were mounted in DAPI mounting medium (Vectashield, Reactolab SA, Servion, Switzerland). A Leica DMI6000B microscope was used for analysis of the specimens.

Statistical analysis
Statistical analysis was performed with graph pad prism version 6 (GraphPad Software Inc., La Jolla, CA, USA). Results are expressed as mean ± standard error of the mean (SEM). Significance was assumed if p<0.05 (F-test) in comparison to the respective controls.