Chromatin remodeling gene AT-rich interactive domain-containing protein 1A suppresses gastric cancer cell proliferation by targeting PIK3CA and PDK1

The tumor suppressor gene AT-rich interactive domain-containing protein 1A (ARID1A) was frequently mutated in cancers. The modulation mechanism of ARID1A for PI3K/AKT signaling in gastric cancer (GC) remains elusive. Here, we found that depletion of endogenous ARID1A enhanced the in vitro proliferation, colony formation, cellular growth, nutrient uptake and in vivo xenograft tumor growth of GC cells. PI3K/AKT activation by ARID1A-silencing was profiled using a phospho-protein antibody array. The phosphorylation of PDK1, AKT, GSK3β and 70S6K, and the protein and mRNA expressions of PI3K and PDK1, were upregulated by ARID1A-silencing. Chromatin immunoprecipitation and luciferase reporter assay revealed that ARID1A-involved SWI/SNF complex inhibited PIK3CA and PDK1 transcription by direct binding to their promoters. Serial deletion mutation analyses revealed that the ARID1A central region containing the HIC1-binding domain, but not the ARID DNA-binding domain and the C-terminal domain, was essential for the inhibition of GC cell growth, PI3K/AKT pathway phosphorylation and its transcriptional modulation activity of PIK3CA and PDK1. The proliferation, cellular growth and glucose consumption of ARID1A-deficient GC cells were efficiently prohibited by allosteric inhibitors mk2206 and LY294002, which targeting AKT and PI3K, respectively. Both inhibitors also downregulated the phosphorylation of PI3K/AKT pathway in ARID1A-deficient GC cells. Such cells were sensitized to the treatment of LY294002, and AT7867, another inhibitor of AKT and p70S6K. The administration of LY294002 alone inhibited the in vivo growth of ARID1A- deficient GC cells in mouse xenograft model. Our study provides a novel insight into the modulatory function and mechanism of ARID1A in PI3K/AKT signaling in GC.

These mutations were prevalent for frameshifts or nonsense mutations, which will lead to mRNA decay, protein miss-folding or domain dysfunction. Loss of ARID1A expression is frequent in a variety of cancers, especially in gynecologic cancers [19,20]. ARID1A/ BAF250a was absent in 51% of primary GCs and was significantly associated with poor prognosis [5,21]. We also found that 24% of GC samples analyzed were ARID1A-negative [22]. However, Kim MS et al argued that loss of ARID1A expression was not common in GC [23]. Wiegand et al found that ARID1A was lost in 20-22.5% of GCs but not significantly associated with any clinical parameters [24]. The intriguing observations emphasize a need for additional analyses.
ARID1A mutation in cancer tended to occur in a synergistic fashion with PIK3CA [5,8,[29][30][31][32]. Silencing of ARID1A in glioma, ovarian and colon cancer cells upregulated the phosphorylation of AKT and P70S6K [33][34][35]. Despite the findings, no further analysis has been performed to get insight into the modulatory mechanism of ARID1A of PI3K/AKT signaling. Given that ARID1A is a transcriptional modulator instead of a protein kinase, the direct targets of ARID1A in PI3K/AKT pathway remains elucidative. Furthermore, the modulation role of AIRD1A in GC needs to be further addressed.
In the present study, we analyzed the ARID1A functions in GC cell proliferation, cellular growth and nutrient consumption in vitro and in vivo. We performed a phosphorylation profiling of PI3K/AKT signaling in GC cells with ARID1A depletion and identified the direct transcriptional targets of ARID1A in PI3K/AKT pathway. We also mapped the essential region of ARID1A protein in the transcriptional regulation of its target genes. We analyzed the in vitro and in vivo drug responses of GC cells with ARID1A depletion.

ARID1A depletion enhances the proliferation and growth of GC cells
We silenced endogenous ARID1A in GC cell lines MGC-803, AGS, HGC-27 and/or SGC-7901 using a siRNA or shRNAs. The siRNA remained as effective till 5 days post-transfection (Supplementary Figure 1). The proliferation of GC cell lines was enhanced comparing with controls, as revealed by MTT or cell counting method ( Figures 1A-1F). The immunofluorescence of Ki-67, a typical nuclear proliferation antigen, was increased in ARID1A-depleted cells (Supplementary Figure 2). HGC-27 and SGC-7901 cells with silenced ARID1A produced more colonies comparing with controls ( Figure 1G and 1H). The average cell sizes ( Figure 1I and 1J, Supplementary Figure 3A and 3B) and the glucose consumptions ( Figure 1K and 1L, Supplementary Figure  3C and 3D) of GC cells and Hela cells were increased significantly after ARID1A knockdown, suggesting that ARID1A depletion speeded up nutrients consumption and cellular growth.
To verify the role of ARID1A in vivo, we performed a xenograft tumor assay in mice. The HGC-27 cells with ARID1A deficiency produced larger (p = 0.01) and heavier (p = 0.04) tumors than the controls with luciferase silencing ( Figure 1M). This in vivo assay strengthen the notion that the single gene depletion of ARID1A would promote tumor formation.

ARID1A knockdown activates PI3K/AKT phosphorylation cascade in GC cells
To date, no analysis was performed to profile the phosphorylation changes of PI3K/AKT signaling in GC. The direct target(s) of ARID1A in PI3K/AKT pathway remains unclear. Here, we profiled the phosphorylation changes of PI3K/AKT pathway in SGC-7901 cells with ARID1A silencing using an AKT/PKB phospho antibody array (Supplementary Figure 4). The phosphorylations of the major components of PI3K/AKT pathway were increased, including AKT, mammalian target of rapamycin (mTOR), glycogen synthase kinase 3α/β(GSK3α/β), p53, p70S6K, PDK1, Bcl-2-associated death promoter (BAD), B-cell lymphoma 2 (BCL-2), tuberous sclerosis complex 2 (TSC2), 14-3-3, p21 and p27 (Figure 2A  and Supplementary Table 3). These results indicated that ARID1A deficiency in GC cells accelerated cell proliferation and nutrient uptake, but inhibited apoptosis on the other hand. The upregulation of p-AKT T308 , p-AKT S473 , p-GSK3 S9 and p-S6K T389 were verified using Western blot ( Figure 2B). Interestingly, the total protein www.impactjournals.com/oncotarget transfected with a siRNA targeting ARID1A after plating for 24 hrs. After a further culture of 24 hrs, the cells were seeded onto a 96-well plate for growth assay. Cell proliferation was measured using MTT method at day 1, 2, 3, 4 and 5. The Western blot images showed the downregulation of ARID1A in the GC cell lines at day 5. NC, negative control of transfection by scramble siRNA. C-F. ARID1A was stably silenced using shRNAs in GC cell lines AGS, SGC-7901 and HGC-27. AR, ARID1A. Cell proliferation was detected using MTT method (C-E) or cell count (F). G, H. Colony formation assay of GC cell lines HGC-27 and SGC-7901 with ARID1A silencing. The colonies were counted manually under a microscopy and the significance was calculated using Student's t test. expressions of PI3K and PDK1 were upregulated by ARID1A-depletion.
The increased phosphorylation of p21 T145 and p27 S10/T187 indicated the dysregulation of cell cycle by ARID1A depletion. ARID1A silencing increased cell percentage in S phase and broke G2 arrest (Supplementary Figures 5A and 5B). Whereas restored expression of wild-type ARID1A in 293FT cell caused G2 arrest and decline of cell percentage in S phase (Supplementary Figure 5C). The expressions of two downstream targets of p53, CDKN1A (encoding p21, a negative regulator of cell cycle) and SMAD3, were decreased in HGC-27 with ARID1A silencing (Supplementary Figures 5D-5F). This observation reinforced previous findings in OCCC that ARID1A might regulate cell cycle in collaboration with p53 [20].
PIK3CA and PDK1 were the direct transcriptional targets of ARID1A ARID1A is not a protein kinase, but rather a chromatin-remodeling protein. So we postulate that ARID1A might directly regulate PIK3CA and/or PDK1 at the transcriptional level. As expected, both PIK3CA and PDK1 genes were increased by ARID1A silencing with a siRNA or shRNA in GC cells or Hela ( Figure 3A and 3B). However, PTEN gene remained unchanged, indicating PTEN might not be a transcriptional target of ARID1A ( Figure 3B).
Dual-luciferase reporter assays of PIK3CA and PDK1 promoters revealed that the luciferase activities of both promoters were increased significantly by  Figure 3C). We then did a Chromatin immunoprecipitation (ChIP)-PCR assay using an antibody against BRM/SWI2-related gene 1 (BRG1), the core catalytic ATPase subunit of SWI/SNF complex [36]. ARID1A-involved SWI/SNF complex directly interacted with PIK3CA promoter within -1127 ~ -946 bp ( Figure 3D). Whereas the binding of the SWI/SNF complex to PDK1 promoter might occur at regions of -1250 to -1154 and -385 to -286 bp ( Figure 3E). These results indicated that PIK3CA and PDK1 were the direct transcriptional targets of ARID1A in PI3K/AKT pathway of GC cells. Thus both PIK3CA and PDK1, as well as p-AKT, were potential drug targets in ARID1Amutant cancer cells.

Hypermethylated in cancer 1 (HIC1)-binding domain of ARID1A protein is essential for the modulation of its downstream transcriptional targets
ARID1A protein contains an AT-rich interactive domain (ARID) (aa 1017-1108), a HIC1-binding domain (aa 1355-1424)/Gln domain (aa 1327-1404) [26], a glucocorticoid receptor (GR)-binding domain (aa 1635- expressions of PIK3CA and PDK1 were analyzed using quantitative reverse transcription PCR (qPCR) after ARID1A was transiently silenced using siRNA. B. The mRNA expressions of of PIK3CA, PDK1 and PTEN were analyzed using qPCR after ARID1A was stably knocked down in HGC-27 and Hela cells using a shRNA. C. HGC-27 cells were stably transfected with pLKO.1 (empty vector) and sh3-ARID1A and luciferase reporter assays were performed to analyze the activities of different promoter constructs of PIK3CA and PDK1, which were depicted to the right side. The arrow indicates the transcription start. The numbers indicate the sites (bp) of 5' terminals of the promoter fragments. D, E. HGC-27 cells were stably transfected with sh-luciferase (sh-Luc) and shRNAs (ARID1A) and ChIP was performed using an antibody against the core catalytic subunit of the SWI/SNF complex, BRG1, or using mouse IgG as a negative control. The PCR amplified regions were illustrated by two opposite arrow heads and the start and end sites were depicted. The box with diagonal lines indicates the first exon of PDK1 gene. www.impactjournals.com/oncotarget 2285) [37] and four leucine-rich steroid receptor binding LXXLL motifs ( Figure 4A). To map the critical region of ARID1A for transcriptional modulation of PIK3CA and PDK1 in GC cells, we produced deletion mutations of ARID1A protein ( Figure 4A). Overexpression of the fulllength ARID1A (ARID1A-fl) significantly downregulated the proliferation of 293FT cells ( Figure 4B) and the levels of PI3K, p-PDK1 S241 /PDK1, p-AKT S473 , p-S6K T389 and p-GSK3β, but increased p21 expression, in AGS cells ( Figure 4C). PTEN had no obvious change by ARID1A-depletion.
ARID1A-C1 overexpression inhibited the proliferations of Hela and AGS ( Figure  4D and 4E) and downregulated PI3K/AKT signaling ( Figure 4F), suggesting the regulation of PI3K pathway by ARID1A was independent of DNA-binding activity of its ARID domain [38]. ARID1A silencing upregulated the activity of PI3K/AKT pathway, while ARID1A-C1 overexpression reduced the activity of the pathway ( Figure 4G). Overexpressions of ARID1A-C2, C3 and C4 did not change PI3K and p-AKT S473 expressions and GC cell proliferations ( Figures 4H-4I). We failed to overexpress the N-terminus of ARID1A (aa 1-1201) in GC cells (data not shown). Luciferase promoter reporter assays revealed that ARID1A-C1 overexpression significantly reduced the activity of promoter construct PIK3CA-p1 illustrated in Figure 3C ( Figure 4J). Similarly, ARID1A-C1 overexpression downregulated the transcriptional activity of PDK-1-p1 promoter ( Figure 3C) in AGS and SGC-7901 cells ( Figure 4K). These results suggested that the region from amino acid 1202 to 1531 of ARID1A, which contains the HIC1binding domain, was essential for the regulation of PI3K/ AKT signaling in GC cells.

MK2206 and LY294002 suppress the proliferation and growth of GC cells with ARID1A depletion by deactivation of PI3K/AKT signaling
The above findings suggested that PI3K, PDK1 or p-AKT might be the therapeutic targets of GCs with ARID1A deficiency. Given that PIK3CA mutation occurs frequently in a synergistic fashion with ARID1A, we focused on the drug effectiveness of small inhibitors against PIK3CA, as well as p-AKT, which is a "hub" target of diverse tumorigenic signaling. Although the drug sensitivity of ARID1A-deficiency has been analyzed in ovarian cancer cell lines [39], no in vivo effectiveness of PI3K/AKT inhibitors for GC cells with ARID1A depletion has been addressed.
ARID1A was silenced in HGC-27 and SGC-7901 cells, which were treated with mk2206, an allosteric inhibitor that inhibits Akt phosphorylation. p-AKT S473 and cell growth were obviously upregulated by ARID1A depletion but were suppressed by mk2206 treatment (Figure 5A and 5B). The average cell volume and glucose consumption of AGS and Hela cells increased by ARID1A silencing, were significantly inhibited by mk2206 (especially for glucose consumption) ( Figures 1I and 1K, Supplementary Figure 3B).
We analyzed the inhibitory activity of LY294002, a selective inhibitor of PI3K, for GC cells with ARID1A deficiency. The cell proliferation ( Figure 5C and 5D), glucose consumption ( Figure 1L and Supplementary Figure 3D), cell size ( Figure 1J and Supplementary Figure 3B) and p-AKT substrate activity ( Figure 5E and Supplementary Figure 3E), were upregulated by ARID1A depletion, but were dramatically reduced by LY294002 treatment. The reduction of nutrient uptake by ARID1A-deficiency was more prominent than cell volume. Consistently, the upregulation of the major components of PI3K/AKT pathway induced by ARID1A silencing, like PI3K, PDK1, p-AKT S473 , p-GSK3 and p-S6K T389 , were inhibited by LY294002 (Figures 5F-5H).
SGC-7901 cells with gene silencing were inoculated into nude mice, which were then treated with LY294002 by intraperitoneally injection after the establishment of tumors ( Figure 6E). Interestingly, established tumors with ARID1A-depletion were significantly shrunk by LY294002 treatment comparing with mock controls after 8 days of administration (p = 0.041 at 8 th day and p = 0.043 at 9 th day) ( Figure 6F). However, no obvious changes were observed between the drug treatment and mock treatment of tumors with native ARID1A.
These observations suggested that mk2206 and LY294002 targeting PI3K/AKT signaling could effectively suppress the proliferation of GC cells with ARID1Adeficiency.

DISCUSSION
ARID1A gene emerges as a bona fide tumor suppressor, while its function and molecular mechanism in GC remains elusive. Although ARID1A has been observed to regulate GC cell proliferation by in vitro experiments [5,21], no in vivo evidence was established to date. Here we verified the tumorigenesis role of ARID1A knockdown  in GC by in vivo xenograft mouse model and in vitro proliferation analyses. Another interesting finding was that ARID1A depletion enhanced cellular nutrient uptake and increased average cell size, suggesting ARID1A deficiency enhanced tumorous metabolism processes other than accelerating cell cycle, suppressing apoptosis, triggering EMT and increasing cancer cell migration/invasion [20,22,27,40].
In current study, we profiled and verified the phosphorylation of PI3K/AKT pathway in GC cells with ARID1A depletion using an antibody array and Western blot. The increased phosphorylations of AKT, www.impactjournals.com/oncotarget PDK1, TSC2, mTORC1 and p70S6K were in line with the enhanced cellular growth of GC cells. The increased phosphorylation of GSK3b was in agreement with the upregulation of glucose consumption, while the phosphorylation of p27 Kip1 and p21 Cip1, two cell cycle inhibitors, were related with accelerated cell cycle. The upregulated phosphorylation of BCL2 were associated with an inhibition of apoptosis [41]. Increased MDM2 S166 indicated a suppression of p53 function [42]. The increased phosphorylation of BAD S134/S136 and 14-3-3 contributed to GC cell survival. These results suggested that the activation of PI3K/AKT pathway by ARID1A depletion contributed to multiple cellular alterations including uncontrolled cell cycle, increased survival ability and enhanced nutrient uptake, which had been addressed with additional experiments in current study. ARID1A is one of the 9 to 12 subunits of SWI/SNF chromatin remodeling complexes which contributed to nucleosome remodeling and transcriptional regulation of a broad profile of target genes [19,36]. Although the upregulation of p-AKT in cancer cells by ARID1A deficiency has been observed, no direct link was established between ARID1A, a transcriptional regulator, and PI3K/AKT pathway, a typical pathway by phosphorylation cascade [31,33,39,43]. In our analysis, we showed evidences that PIK3CA and PDK1 were the direct transcriptional targets of the ARID1Ainvolved SWI/SNF complex. Furthermore, we found that the negative regulation function of ARID1A for PIK3CA and PDK1 was independent of its N-terminus containing the ARID domain, the LXXLL motifs and its C-terminal fragment containing the GR-binding domain, but interestingly relied on the central region containing the HIC1-interaction domain. ARID domain was known to mediate the binding of ARID1A to DNA sequence in a non-specific fashion [38]. Nie Z et al identified that GRdependent transcriptional activation was reduced slightly by the deletion of the ARID DNA-binding domain but was dramatically reduced by the deletion of its C-terminal GR-binding domain (aa 1637 to 2285) [37]. Here, we showed further evidence that not only the ARID domain but also the GR-binding domain might not have dominant role for its transcription activity, at least in PI3K/AKT signaling in GC cells. Guan B et al reported that the C-terminus (aa 1759 to 2285) of ARID1A, the same fragment as ARID1A-C3 in our study, interacted with p53 to regulate the transcription of p21, a downstream target gene of p53 [20]. Although the N-or C-terminus of ARID1A was analyzed for its interaction with other proteins, its function in cellular proliferation or growth has not been tested. For the first time, we showed that the C-terminus of ARID1A (1202-2285, ARID1A-C1), but not the shorter forms as C2 (1531-2285), C3 (1759-2285) and C4 (1867-2285), had suppression activity for GC cell proliferation. C3 preserved the activity to bind to p53 as established in 293FT cells by Guan B et al [20], however, its inability in GC cell growth inhibition indicated that the binding of ARID1A to p53 might not be essential for its transcriptional activity for target genes other than the genes downstream of p53. HIC1 is a tumor suppressor and a transcriptional repressor [44], which inhibits E2F1 transcription through the recruitment of SWI/SNF complex by a direct interaction with ARID1A [26]. ARID1A/BAF250a might function as a scaffolding factor through protein-protein interaction that recruiting transcriptional suppressors like HIC1 to promoters of the targeted genes.
As described above, the transcriptional mechanism of ARID1A would be addressed to deal with the heterogeneity of different cancers and physiological and pathological variations. Although ovarian cancer cell lines with ARID1A-deficiency showed some in vitro sensitivities to PI3K/AKT inhibitors [39], no in vivo responsiveness of GC cells with ARID1A-deficiency effectiveness of PI3K/AKT inhibitors has been addressed. We analyzed the effects of chemical inhibitors against PI3K/AKT signaling in vitro and in vivo. Mk2206 and LY294002, which respectively targeting p-AKT S473 and PI3K, were effective to downregulate GC cell proliferation, glucose consumption and cellular growth (cell size) conferred by ARID1A depletion. Accordingly, the PI3K/AKT signaling was downregulated by mk2206 or LY294002. The drug sensitivity of ARID1A-deficient GC cells was increased for the treatment of LY294002 and AT7867, another inhibitor against AKT and p70S6K. The administration of LY294002 alone inhibited the in vivo growth of the xenograft tumors with ARID1A silencing, comparing with the cells with native ARID1A.
Taken together, our study revealed that deficient ARID1A confers an increased proliferation, cellular growth and nutrient consumption of GC cells, companied with an activation of PI3K/AKT signaling. PIK3CA and PDK1 were the direct transcriptional targets of the ARID1A-involved SWI/SNF complex. The central region containing the HIC1-binding domain of ARID1A was essential for its transcriptional regulation of downstream targets. ARID1A deficiency and the activation of PI3K/ AKT were hazard factors of GC patients and associated with poor prognosis. MK2206 and LY294002 were effective to suppress the proliferation of GC cells with ARID1A deficiency by downregulating activated PI3K/ AKT signaling. Our study provides a novel insight into the function and mechanism of ARID1A in the modulation of PI3K/AKT signaling in GC.

Cell lines
GC cell lines HGC-27 and MGC-803 were purchased from the Cell Bank of Shanghai Institutes for Biological Sciences, China. SGC-7901 was generously provided by Dr. Jian-Jun Du at Huashan Hospital, Fudan University. AGS was a gift by Dr. Qing-Hua Zhang at the Shanghai University of Traditional Chinese Medicine. GC cells were cultured in RPM1640 media supplemented with 10% FBS in a humidified incubator at 37°C and 5% CO 2 . Hela and HEK293FT were cultured in DMEM supplemented with 10% FBS. The cell lines were authenticated on August, 2014.

Gene cloning and lentivirus production
The ARID1A gene was cloned as described previously [22]. The shRNA lentiviral plasmid (pLKO.1) was purchased from the RNAi consortium. The shRNAs of ARID1A had been described previously [22]. Lentivirus was generated in HEK293FT cells using the second-generation packaging system pMD2.G (Addgene plasmid 12259) and psPAX2 (Addgene plasmid 12260). Lentiviral titer was determined as described [45]. The primers and shRNAs used were listed in Supplementary Table 1.

Western blot analysis
Immunoblotting analysis was performed as previously described [22]. All antibodies used in current study were listed in Supplementary Table 2.

Immunofluorescence
Immunofluorescence was performed to monitor the proliferation of GC cells after ARID1A silencing according to the procedure established previously [46].

Phospho-protein antibody array
The experiment was performed by Wayen Biotechnologies (Shanghai), Inc., using the Full Moon AKT/PKB phospho antibody array containing 216 antibodies against the corresponding proteins of AKT/ PKB pathway (Full Moon BioSystems, Inc.). There are 93 paired antibodies, which react with a phosphorylated site and a paired non-phosphorylated site. Proteins were extracted from SGC-7901 or HGC-27 cells stably transfected with sh-Luciferase or sh2 (ARID1A) using the Extraction Buffer (Full Moon BioSystems, Inc.). Next, 25 µg proteins of each sample were labeled with 30 µg biotin in 4% N,N-Dimethylformamide solution. The reaction was stopped after 2 hours of incubation at room temperature. The labeled proteins were diluted with 6 mL Coupling Solution (Full Moon BioSystems, Inc.). The antibody arrays were blocked with the Blocking Reagent (Full Moon BioSystems, Inc.) at room temperature for 45 min and then rinsed 10 times with Milli-Q water. The diluted labeling products were incubated with the antibody arrays at 4°C overnight. The array slides were washed with Milli-Q water as above. The conjugated proteins on the array slides were detected with Cy3-streptavidin (0.5 µg/mL) solution. The slides were scanned using a microarray scanner GenePix 4000B with software GenePix Pro 6.0 (Axon Instruments, USA). The raw data was treated with the Grubbs' algorism and outliers in each six-repeat data point of an antibody were excluded. The data was average for each antibody and all data were normalized using the average against beta-actin. The phosphorylation ratio = phospho/unphospho. The phosphorylation fold change = (phosphorylation sh2-ARID1A)/( phosphorylation sh-Luciferase). The proteins with a fold change ≥ 1.8 were reported in Supplementary  Table 3. The differential significance were evaluated with two-sided Student's t test and a p value < 0.05 was considered as statistically significant.

Cell cycle assay
Cell cycle alteration by ARID1A silencing was measured using the Cell Cycle Detection Kit (Biyuntian Inc. Haimen, China) with a Beckman Flow Cytometer (Beckman Coulter, Inc.). Cells in suspension were fixed by adding 2.5 mL 70% alcohol and incubated at 4°C overnight. Alcohol was aspirated with centrifugation and the cells were washed with PBS/1% BSA triple times and suspended with 50 µg/mL propidium iodide and 0.1 mg/ mL RNase. After incubation at 37°C for 40 min, the cells were washed by adding 3 mL PBS/1% BSA. Finally, the cells were re-suspended with 1 mL PBS and analyzed by a cytometer.

RNA extraction and qPCR
Total RNA isolation and qPCR were performed as previously described [22]. Primers are listed in Supplementary Table 1.

Colony formation
Colony formation assay was performed as describe previously [47].

Cell transfection
Plasmid or siRNA transfection using lipofectamine (Life Technologies -Invitrogen) and OPTI-MEM reduced serum media (Life Technologies -Invitrogen) was performed as described previously [46]. The cells were transfected after plating for 24 h. After a further incubation of 24 h, the cells were seeded onto a 96-well plate for growth curve analysis. The sequences of ARID1A siRNAs were synthesized as previously reported [22]. The siRNAs were listed in Supplementary Table 1.

Glucose consumption
Cells were seeded in culture dishes and the media were changed after 6 h. The cells were further incubated for 24 h, and the culture media were collected and used for glucose measurement using a Glucose Assay kit (Sigma-Aldrich, Shanghai, China). The assay was performed according to the manufacturer's instructions. The glucose consumption and lactate production were normalized by cell counts (per 10 6 ). First, 1 mL of culture medium of each sample was added into a test tube and a standard sample in another tube was prepared by adding 950 µL of water and 50 µL of the Glucose Standard solution into the tube. The reactions were started at zero time by adding 2 mL of the Assay Reagent to each tube at an interval of 30-60 seconds. The tubes were incubated at 37°C for 30 min. The reactions were stopped by adding 2 mL of 12 N H 2 SO 4 into each tube at 30-60 seconds intervals. The absorbance of each tube against the reagent blank at 540 nm was measured using a microplate spectrophotometer. The glucose concentration was calculated as: mg glucose = (ΔA 540 of Test) (mg glucose in Standard)/ΔA 540 of Standard. The glucose consumption was: (original glucose concentration in the medium) -(glucose concentration at test).

Growth curve assay
Cell proliferation was measured using methylthiazolyldiphenyl-tetrazolium bromide (MTT) assay as previously described [46]. Alternatively, cells were counted by a Vi-Cell XR cell viability analyzer (Beckman Coulter). Inhibition assays were performed by adding 1 nmol/L mk2206 dihydrochloride (Selleck Chemicals, Shanghai, China) or 10 or 25 µM LY294002 (Selleck Chemicals, Shanghai, China) to the culture medium.

Luciferase reporter assay
The promoter regions of PIK3CA and PDK1 was amplified using normal human genomic DNA, and inserted into pGL3 luciferase reporter vector (Promega, Shanghai, China) [22]. Three promoter regions of PIK3CA were cloned, including -1889 to +1 (transcriptional start), -1276 to +1 and -516 to +1. The promoter region from -1325 to +1 (transcriptional start) of PDK1 was cloned for analysis. The primers were deposited in Supplementary  Table 1. The procedure has been described previously [22].

ChIP assay
ChIP assay was performed using an antibody against BRG1 or IgG as a negative control with an EZ-ChIP kit (Upstate Biotechnology) according to the manufacturer's instruction and previous reports [22,48]. The primers used to amplify the promoter fragments were listed in Supplementary Table 1.

Drug sensitivity assay
Gastric cancer cells (1-2 × 10 4 per well) were seeded on a 96-well plate and treated with different concentrations of drug chemicals for 72 hrs at 37°C. The concentrations of LY294002 and AT7867 (Selleck Chemicals, Shanghai, China) were 0, 0.01, 0.1, 1, 10 and 100 µM. Twenty µL of MTT solution (5 mg/mL) (Sigma-Aldrich, Shanghai, China) was added to each well and the cells were incubated for 4 hrs at 37°C. Finally, 150 µL DMSO was added and allowed to dissolve for 10 min. The absorbance was monitored at 490 nm using a microplate spectrophotometer.

Xenograft mouse transplantation model
The BK nude mice (male) were 4-5 weeks old at the beginning of the study. These mice were purchased from Shanghai Laboratory Animal Center, Shanghai, China, and allowed to acclimate for 5 days in the animal facility before further intervention. Cells (5 × 10 6 ) mixed with equal volume of Matrigel (BD Biosciences, Shanghai, China) were injected subcutaneously into both flanks of each mouse. When the tumor volume reached 100 mm 3 , the mice were injected intraperitoneally of LY294002 (Selleck Chemicals, Shanghai, China) every day with a dosage of 30 mg/kg (body weight). The control group was injected with 1 × PBS (pH 7.4). Tumor diameters were measured using a caliper every two days and the tumor volume was calculated using the formula: V (mm 3 ) = (width) 2 × length/2. The mice were executed 30 days after cell inoculation. The tumors were isolated from the bodies and weighed, followed by fixation with 4% neutral paraformaldehyde. All experimental procedures were conducted according to the guidelines of the Animal Experimental Ethics Committee of Shanghai Medical College of Fudan University.

Statistics
Differential significance was calculated using unpaired Student's t test. Drug sensitivity statistics was performed using GraphPad Prism 6 and the concentration was Log transformed and the divergence of LogIC50 was evaluated with an F-test. A p value < 0.05 was considered as statistically significant. All statistical tests were twosided.

ACKNOWLEDGMENTS
This work was supported by Chinese National Key Program on Basic Research (973) (2011CB910702 and