Forkhead Box F2 Suppresses Gastric Cancer through a Novel FOXF2–IRF2BPL–β-Catenin Signaling Axis

Gastric cancer (GC) is one of the most common cancers all over the world, especially in Asia (1). The prognosis of gastric cancer patients is still poor, in spite of the improved surgical and adjuvant treatment approaches. Five-year overall survival of gastric cancer is generally 25%–30% (2). There are accumulating evidences that aberrant DNA methylation is a hallmark of gastric cancer (3). Identification of novel tumor suppressor genes repressed by DNA promoter methylation will provide new insights into the molecular pathogenesis of gastric cancer, and will be useful in discovering potential biomarkers for diagnosis and treatment of this disease (4). Using the promoter methylation array for a genome-wide screen of hypermethylated candidates in gastric cancer, we identified that the promoter of Forkhead box F2 (FOXF2) gene was preferentially methylated in gastric cancer cell lines. FOXF2 was first identified to be frequently rmethylated in childhood acute lymphoblastic leukemia, colorectal cancer, and kidney cancer (5). However FOXF2 methylation status in gastric cancer has not been studied yet.

FOXF2, located on chromosome 6p25 and encoded a 444 amino-acid protein, belongs to the forkhead family of transcriptional regulators (6, 7). It has been described as an essential signaling molecule for embryogenesis and tissue development (8–11). Recently, decreased FOXF2 was shown in certain human cancers of prostate, breast, liver, and esophageal squamous cell carcinoma (12–15). However, the role of FOXF2 in human gastric cancer is still unknown. In this study, we investigated functional significance, molecular mechanisms, and clinical impact of FOXF2 in gastric cancer.

Forty paired primary gastric cancer and adjacent nontumor samples were obtained during operation of gastric cancer patients diagnosed in the Prince of Wales Hospital of the Chinese University of Hong Kong from 2015 to 2017. In addition, 103 primary gastric tumors DNA were obtained from Zhejiang University, China. We used the same inclusion and exclusion criteria for both cohorts. Gastric adenocarcinoma patients with age >18 were enrolled in this study. Pregnant or nursing patients were excluded. Written informed consents were obtained from subjects or their authorized representatives. The study protocol was approved by the Ethics Committee of the Chinese University of Hong Kong and the Ethics Committee of Zhejiang University. This study was carried out in accordance with the Declaration of Helsinki of the World Medical Association.

Gastric cancer cell lines (AGS, HGC27, MKN1, MKN28, MKN45, MKN74, NCI-N87, SNU719), L cell, L Wnt-3a cell were purchased from the ATCC and MGC803 was purchased from Chinese Academy of Sciences Cell Bank (CAS Cell Bank, Beijing, China). 293ft cell line was purchased from Invitrogen (Thermo Fisher Scientific). Cell lines were maintained according to the protocols from ATCC. All cell lines were obtained between 2013 and 2015 and cell identities were confirmed by short tandem repeat profiling. Routine Mycoplasma testing was performed by PCR. Cells were grown for no more than 25 passages in total for any experiment.

Genomic DNA samples isolated from three gastric cancer cell lines (AGS, MKN45, MGC803), normal gastric cell line GES1, and two normal stomach tissues were employed for promoter methylation analysis by the Infinium HumanMethylation450 BeadChip array (Illumina).

To externally validate the epigenetic regulation of FOXF2 gene expression in gastric cancer, we analyzed publicly available datasets of methylome profiling on the Infinium DNA Methylation Array [Stomach Adenocarcinoma (STAD) Methylation450k and Methylation27k datasets] and transcriptome sequencing on the Illumina HiSeq 2000 platform provided by The Cancer Genome Atlas (TCGA) multicenter project. IDAT files and raw mRNA sequence counts were obtained from the TCGA Data Portal (https://gdc.cancer.gov/).

DNA BGS analysis was performed as described previously (16). Ten CpG sites spanning from −1037 to −836 bp relative to the transcription start site (TSS) were evaluated. Primer sequences for BGS are listed in Supplementary Table S1.

Cell lines were treated with demethylation reagent 5-aza-2′-deoxycytidine (5-Aza; 2 μmol/L, Sigma-Aldrich) for 5 days. The 5-Aza was replenished every day. Some cell lines were further treated with histone deacetylase inhibitor Trichostatin A (TSA, 300 nmol/L) for additional 24 hours. Cells were treated with cell-permeable proteasome inhibitor MG132 (Millipore) and chloroquine (Sigma-Aldrich) for 12 hours before harvesting for Western blot or immunoprecipitation assays.

Human FOXF2 (NM_001452) ORF Clone (RC212020) was purchased from Origene. The deletion of FOXF2 (ΔFOXF2) (N-terminal deletion of residues 1–200) was amplified by PCR. Primer sequences are listed in Supplementary Table S1. Stable transfections were selected for 1 week with G418 antibiotic.

Cells were transfected with FOXF2 siRNA (RiboBio) or control siRNA using Lipofectamine 2000 (Thermo Fisher Scientific). The sequences of the siRNA used are listed in Supplementary Table S1.

Semiquantitative PCR was performed by AmpliTaq Gold DNA polymerase (Applied Biosystems; Thermo Fisher Scientific). Quantitative real-time PCR was performed by SYBR Green PCR Master Mix (Applied Biosystems; Thermo Fisher Scientific) on 7500HT Fast Real-Time PCR System (Applied Biosystems; Thermo Fisher Scientific). Primer sequences are listed in Supplementary Table S1.

Proteins were separated on 12% SDS-PAGE and transferred onto nitrocellulose membranes (GE Healthcare). Blots were immunostained with primary antibody and secondary antibody. Independent experiments were performed at least twice. The antibodies used and their dilutions are listed in Supplementary Table S2.

After 24 hours of transfection, cells were seeded for 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium (MTS; Promega) assay to measure the cell viability. Cells were seeded on 6-well plates. After five days, cells were fixed with 70% ethanol for 10 minutes and stained with 0.5% crystal violet solution for 10 minutes. Colony with more than 50 cells per colony was counted. The experiment was conducted in three independent triplicates.

Transfected cells were fixed in 70% ethanol and stained with PI/RNase Staining Buffer (BD Biosciences). The cells were sorted by FACSCalibur system (BD Biosciences) and analyzed by Flowjo 7.6 software (BD Biosciences).Transfected cells were stained with Annexin V-APC (BD Biosciences) and 7-aminoactinomycin (7-AAD). The cells were sorted by FACSCalibur system and Annexin V–positive cells were counted as apoptotic cells.

Cells were transfected with the indicated plasmids and stained with primary antibody and secondary antibody. Images were captured by fluorescent microscopy. The antibodies used and their dilutions are listed in Supplementary Table S2.

Cells were chemically cross-linked by formaldehyde and sonicated at 4°C. Cell was then cleared by centrifugation and supernatant was incubated overnight at 4°C with 100 μL of Protein G magnetic beads that had been preincubated with 10 μg of the appropriate antibody for at least 3 hours. Bound complexes were eluted from the beads by heating at 65°C and cross-linking was reversed by overnight incubation at 65°C. Immunoprecipitated DNA was purified and used for gene-specific PCR.

Cells in a 24-well plate were cotransfected with luciferase reporter plasmid, and pRL-TK control vector by Lipofectamine 2000. Plasmids of FOXF2, ΔFOXF2, β-catenin wild-type (WT), β-catenin (S33Y), WNT1, or empty vector were cotransfected as indicated per well. After 24 hours of transfection, luciferase activities were analyzed by Dual-Luciferase Reporter Assay System (Promega) and normalized to the control Renilla.

MKN45 cells (1 × 10⁷ cells in 0.1-mL PBS) stably transfected with FOXF2 expression vector or empty vector were injected subcutaneously into the dorsal right flank of 4-week-old female Balb/c nude mice (n = 5 per group). Tumor diameter was measured every 3 days for 3 weeks. Tumor volume (mm³) was estimated by measuring the longest and shortest diameters of the tumor and calculating as described previously (16). The excised tissues were either fixed in 10% neutral-buffered formalin or snap frozen in liquid nitrogen. Tumor sections from paraffin-embedded blocks were used for histologic examination. All experimental procedures were approved by the Animal Ethics Committee of the Chinese University of Hong Kong.

Values are expressed as mean ± SD. The independent Student t test was used to compare the difference between two groups. One-way ANOVA was used to compare the difference between multiple groups. Correlation analysis (Spearman ρ) was used to measure the strength of association between FOXF2 methylation and expression. The χ² test was used to compare the clinicopathologic characteristics of gastric cancer patients and FOXF2 methylation. Univariate and multivariate Cox regression analysis was performed to assess the prognostic value of FOXF2 methylation. Overall survival in relation to methylation status was evaluated by Kaplan–Meier survival curve and log-rank test. Differences with P < 0.05 were considered to be statistically significant.

We first examined the FOXF2 expression in normal human tissues and found that FOXF2 was predominantly expressed in gastrointestinal tract including stomach (Supplementary Fig. S1). Compared with its readily expression in normal gastric tissues, FOXF2 was silenced in 6 of 9 gastric cancer cell lines (Fig. 1A). To confirm whether FOXF2 expression was repressed by promoter methylation, methylation status of FOXF2 was evaluated by bisulfite genome sequencing (BGS) analysis in the CpG island of the promoter region covering 10 CpG sites from −1037 to −836 of the FOXF2 gene (Fig. 1B). Full or partial methylation was detected in six gastric cancer cell lines (AGS, MKN28, MKN45, MKN74, NCI-N87, and SNU719), which showed silencing of FOXF2, whereas relatively less methylation was detected in HCG27, MGC803, and MKN1, which showed FOXF2 expression (Fig. 1B). The BGS product covers two CpG probes of 450K array: cg06005891 and cg03848675. These two probes locate at the promoter region of FOXF2 within −1500 bp upstream of transcription start site (Supplementary Fig. S2A). We also evaluated the correlation between 450K array and bisulfite sequencing data. The bisulfite sequencing result was positively correlated with two CpG probe values (Spearman ρ = 0.66, P < 0.05, Supplementary Fig. S2B; Supplementary Table S3). In addition, the promoter methylation level was significantly higher in gastric cancer cell lines compared with normal gastric tissues (P < 0.05; Fig. 1B). In addition, treatment with demethylation agent 5-Aza and histone deacetylase inhibitor TSA restored FOXF2 expression in all gastric cancer cell lines (Fig. 1C).

We next analyzed FOXF2 expression in paired gastric cancer and adjacent normal tissues. FOXF2 mRNA expression was significantly downregulated in 40 gastric cancer tumors as compared with their adjacent normal tissues by real-time PCR (Fig. 1D, left). The downregulation of FOXF2 was validated independently in TCGA cohort (Fig. 1D, right). Moreover, FOXF2 expression was negatively correlated with promoter methylation of gastric cancer tissues in Hong Kong cohort (N = 20, Spearman ρ = −0.57, P = 0.0068) and in TCGA Methylation450k dataset (N = 348, average value of probe cg06005891 and cg03848675, Spearman ρ = −0.28, P = 1.7 × 10⁻⁷; Fig. 1E). Starburst plot integrating alterations in DNA methylation and gene expression showed that FOXF2 was hypermethylated and downregulated in gastric cancer tissues from TCGA datasets (cg03848675, Methylation450k and Methylation27k datasets, Supplementary Fig. S3). Consistently, FOXF2 protein was reduced in gastric cancer tissues compared with adjacent normal tissues as determined by Western blot analysis (P < 0.001). Proliferating cell nuclear antigen (PCNA) protein was served as a marker of tumor tissues in the Western blot analysis (Fig. 1F). We further evaluated FOXF2 in 10 pairs of gastric cancer clinical samples by IHC. FOXF2 was readily expressed in normal gastric tissues but was reduced in cancer counterpart (P < 0.01; Fig. 1G).

The frequent silencing of FOXF2 in gastric cancer cell lines and tissues suggests that FOXF2 may have a tumor-suppressive function. To test this hypothesis, we examined the effect of FOXF2 overexpression on the cell growth in AGS and MKN45, which showed complete silencing of FOXF2. Reexpression of FOXF2 was confirmed by semiquantitative PCR and Western blot analysis (Fig. 2A). FOXF2 markedly reduced cell viability and colony formation ability in AGS (P < 0.01) and MKN45 (P < 0.01) as compared with empty vector (Fig. 2A). We further investigated the effect of FOXF2 knockdown by two FOXF2 siRNAs in HCG27 and MGC803, which have relatively high FOXF2 expression. Silencing of FOXF2 at mRNA and protein level in HCG27 and MGC803 was confirmed by semiquantitative PCR and Western blot analysis (Fig. 2B). Knockdown of FOXF2 significantly enhanced cell viability (P < 0.01) and increased colony numbers in both cell lines (P < 0.01; Fig. 2B).

We next investigated the effect of FOXF2 on migration and invasion abilities of gastric cancer cells using in vitro transwell with or without Matrigel matrix layer. Ectopic expression of FOXF2 in AGS and MKN45 significantly suppressed cell migration and invasive capabilities in both cell lines (Fig. 2C). In keeping with this, Western blot analyses showed that FOXF2 regulated the epithelial–mesenchymal transition (EMT) through upregulation of epithelial markers (E-cadherin) and downregulation of mesenchymal markers (N-cadherin and Vimentin; Fig. 2D). AGS cells are negative for E-cadherin due to the presence of a truncating mutation (17). In contrast, silencing of FOXF2 in HGC27 and MGC803 cells significantly promoted cell migration and invasion (Fig. 2C), and suppressed epithelial markers (E-cadherin), increased mesenchymal markers (N-cadherin and Vimentin; Fig. 2D). These findings indicate that FOXF2 suppressed the migration and invasion ability of gastric cancer.

In light of our in vitro findings, we examined the in vivo tumor-suppressive ability of FOXF2. Empty vector–transfected and FOXF2-transfected MKN45 cells were injected into the right flanks of nude mice, respectively. FOXF2 significantly suppressed the growth of MKN45 xenografts in nude mice (Fig. 2E). The average tumor weight of MKN45-FOXF2 tumors was reduced by 59% compared with controls (0.179 ± 0.136 g vs. 0.434 ± 0.068 g, P < 0.01, Fig. 2E). FOXF2 expression in the xenografts was confirmed by quantitative real-time PCR and Western blot analysis (Fig. 2F). FOXF2-expressing xenografts displayed significantly reduced proliferative activity compared with controls by Ki-67 immunostaining (Fig. 2F). Taken together, our results suggest that FOXF2 inhibits gastric cancer cell growth and functions as tumor suppressor in gastric cancer cells.

We next investigated the effect of FOXF2 on cell-cycle distribution and apoptosis regulation by flow cytometry. FOXF2 significantly increased G₁ phase cells in AGS (64.1% ± 1.0% vs. 69.0% ± 2.1%, P < 0.05) and MKN45 (54.5% ± 0.3% vs. 63.0% ± 3.2%, P < 0.01) but decreased S-phase cells in AGS (13.2% ± 1.5% vs. 8.2% ± 2.5%, P < 0.05) and in MKN45 (13.7% ± 0.5% vs. 8.7% ± 1.1%, P < 0.01; Fig. 2G). In keeping with this, FOXF2 reduced the protein expression of the G₁–S transition promoter cyclin D1, CDK4, and PCNA but enhanced the G₁ gatekeepers p27 and p21. In addition, ectopic expression of FOXF2 induced cell apoptosis in AGS (2.73% ± 0.5% vs. 45.9% ± 8.2%, P < 0.01) and MKN45 (35.7% ± 1.2% vs. 43.1% ± 2.7%, P < 0.05) by flow cytometry following Annexin V-APC and 7-aminoactinomycin (7-AAD) double staining (Fig. 2H). This was confirmed by enhanced the protein levels of key cell apoptosis regulators cleaved caspase-3, cleaved caspase-7, and cleaved PARP (Fig. 2I). These findings suggest that apoptosis in conjunction with cell-cycle arrest, as induced by FOXF2, accounts for the growth inhibition in FOXF2-expressing tumor cells.

FOXF2 is a transcription factor that contains a bipartite nuclear localization signal (NLS) within a DNA-binding domain known as the forkhead domain (18). To confirm the localization and molecular mechanism of FOXF2, we generated the deletion of FOXF2 (ΔFOXF2), which lacks the 2 NLS in the N-terminal (amino acids 1–200; Fig. 3A). Western blot and immunocytochemistry showed that wild-type FOXF2 protein was located in the nucleus, whereas ΔFOXF2 localized exclusively in cytoplasm (Fig. 3A).

To understand the molecular basis of the tumor suppressive property of FOXF2, we performed luciferase reporter screening assays to assess the effect of FOXF2 on five signaling pathways including Wnt, p53/p21, STAT3, AP-1, and MAPK/ERK. Ectopic expression of FOXF2 significantly repressed the activity of Wnt signaling as demonstrated by TOPflash luciferase reporter (Fig. 3B), but not other pathways (Supplementary Fig. S4). The inhibition effect of wild-type FOXF2 on TOPflash reporter became more dramatically in the presence of WNT1 stimulation in AGS and MKN45 (Fig. 3B). However, deletion of the forkhead domain of FOXF2 (ΔFOXF2) abolished the suppressive effect on TOPflash reporter in AGS (Fig. 3B), suggesting the nuclear localization of FOXF2 was functional critical. Conversely, FOXF2 knockdown by two siRNAs in HCG27 and MGC803 increased the reporter activity stimulated by WNT1 and β-catenin, respectively (Fig. 3B). Previous studies indicated that hypereactivated Wnt signaling is critical for gastric cancer cell growth (19, 20). We also found that Porcn inhibitor IWP2 significantly inhibited cell growth in four gastric cancer cell lines (AGS, MKN45, HGC27, and MGC803) with IC₅₀ 3–4 μmol/L by MTT assay. On the other hand, we measured the gastric cancer cell growth treated by Wnt-3A conditioned medium (CM). Treatment of Wnt-3A CM significantly promoted the cell growth of AGS, MKN45, HGC27, and MGC803 compared with the control group (Supplementary Fig. S5A and S5B). Moreover, FOXF2 significantly inhibited Wnt target genes cyclin D1 and c-myc in both mRNA and protein levels, but not change β-catenin mRNA level (Fig. 3C), suggesting FOXF2 negatively regulated Wnt signaling pathway in gastric cancer cells at posttranscriptional level. In addition, we performed Wnt signaling pathway RT² Profiler PCR array and identified a group of Wnt target genes that were also downregulated in gastric cancer cell line AGS upon FOXF2 overexpression (Supplementary Fig. S6). We further validated these genes in AGS and MKN45 by real-time PCR upon stimulated with Wnt3a conditioned medium or L control medium. Overexpression of FOXF2 significantly suppressed mRNA expression of Wnt target genes (Axin2, c-myc, cyclin D1, LEF1, and MET). Treatment with Wnt3a for 2 and 4 hours significantly stimulated the expression of these Wnt targets, whereas these gene inductions were abrogated by FOXF2, suggesting that these genes are direct Wnt targets regulated by FOXF2 (Fig. 3D). To prove the effect of FOXF2 is LEF1 dependent, we constructed an N terminal LEF1 deletion (Δ amino acids 2–65). LEF1(ΔN) loses the β-catenin binding site but well preserves the DNA binding sequencing and therefore it cannot response to Wnt upstream signaling. TOPflash luciferase activity indicated that FOXF2 significantly suppressed TOPflash reporter activity, whereas overexpression of LEF1(ΔN) relieved this effect (Supplementary Fig. S7).

Given that FOXF2 suppressed Wnt downstream targets but failed to cause any changes on β-catenin at mRNA level, we next investigated whether FOXF2 regulated the abundance of β-catenin protein. Ectopic expression of wild-type FOXF2 dose-dependently decreased total β-catenin protein levels in AGS and 293ft cells, whereas ΔFOXF2 had little effect on β-catenin protein (Fig. 4A). We also observed that FOXF2 significantly suppressed β-catenin protein in MKN45, whereas silencing of FOXF2 in HGC27 and MG803 upregulated β-catenin protein (Fig. 2D). Given that FOXF2 suppressed Wnt signaling activity, we next examined whether FOXF2 could interrupt the interaction between β-catenin and its cofactor LEF1 in the nucleus. We coexpressed Flag-FOXF2, Flag-β-catenin, and hemagglutinin (HA)-LEF1 in 293ft cells, and immunoprecipitated HA-LEF1 in cell nuclear lysates with anti-HA antibodies. Western blot analysis of the precipitates with an anti-FLAG antibody indicated that FOXF2 reduced nuclear β-catenin binding with LEF1 protein (Fig. 4B). We next analyzed the membrane, cytosolic, and nuclear amount of endogenous β-catenin in FOXF2 transfected AGS and 293ft cells. Compared with the control cell lines, the membrane amount of β-catenin was unchanged, whereas the cytosolic and nuclear amount of β-catenin was substantially decreased in the FOXF2-transfected cell lines (Fig. 4C). Moreover, immunofluorescence staining also showed that when cells were cotransfected with FOXF2 and β-catenin, FOXF2 reduced the β-catenin protein compared with untransfected cells (Supplementary Fig. S8).

Given that FOXF2 suppressed β-catenin at protein level rather than at mRNA level, we then assessed whether FOXF2 regulated the β-catenin protein stability. We measured the half-life of Flag-β-catenin following inhibition of new protein synthesis by cyclohexamide (CHX). Indeed, the half-life of β-catenin was dramatically decreased in FOXF2-transfected cells within 3 hours compared with the control cells (Fig. 4D). The ubiquitin–proteasome system and autophagy are two major mechanisms that mediate protein degradation. To further determine the molecular mechanism of FOXF2-induced downregulation of β-catenin, we detected β-catenin protein level in the absence or presence of the proteasome inhibitor MG132 and autophagy inhibitor chloroquine. We treated AGS and 293ft cells stably expressing FOXF2 or control vector with MG132 (5 μmol/L and 10 μmol/L) and chloroquine (15 μmol/L and 30 μmol/L) for 12 hours, respectively. MG132 blocked FOXF2-induced degradation of β-catenin. In contrast, we did not observe restoration of β-catenin in cells treated with chloroquine (Fig. 4E). These results suggested that FOXF2-induced degradation of β-catenin is mediated mainly through a proteasome pathway.

To further support the idea that FOXF2 induced β-catenin degradation via an ubiquitin–proteasome pathway, we transiently transfected cells with FOXF2 along with HA-tagged wild-type ubiquitin or control vector. As speculated, FOXF2 increased the level of ubiquitination of β-catenin detected by immunoblot after immunoprecipitation of ubiquitin (Fig. 4F), which suggest that FOXF2 leads to β-catenin polyubiquitylation.

In the canonical Wnt pathway, phosphorylation of β-catenin at the N-terminal aminol acids (Thr41, Ser37, and Ser33) by GSK-3β results in β-catenin proteasome-dependent degradation. We then asked whether FOXF2-promoted β-catenin degradation required GSK-3β. We performed the TOPflash luciferase reporter assay in the presence of wild-type β-catenin or the constitutively active β-catenin S33Y mutant (β-catenin S33Y), which is insensitive to GSK-3β–mediated phosphorylation and subsequent proteasomal degradation. The transcriptional activities of both wild-type β-catenin and β-catenin S33Y were significantly suppressed by FOXF2 (Fig. 4G), suggesting that FOXF2 induced GSK-3β–independent degradation of β-catenin. This was confirmed by Western blot analysis that β-catenin S33Y was degraded by FOXF2 dose-dependently (Fig. 4H, left). Consistent with this observation, β-catenin (Δ1–133), a stabilized form of β-catenin owing to deletion degradation domain in the N terminal, was significantly reduced in the presence of FOXF2 (Fig. 4H, right). To further validate the observation that FOXF2 degraded β-catenin independent of GSK-3β activity, GSK-3β was silenced by two siRNAs (Supplementary Fig. S9A). Knockdown of GSK-3β did not change the effect of FOXF2 on β-catenin protein degradation in AGS and 293ft (Supplementary Fig. S9A). Similar results were observed when cells were treated with selective GSK3 inhibitor Lithium chloride (LiCl; Supplementary Fig. S9B). Collectively, these findings suggested that FOXF2-induced β-catenin degradation is independent of GSK-3β.

Given FOXF2 is a transcriptional factor and lacks of the ability to directly ubiquitinize β-catenin, we hypothesize that FOXF2 transcriptional upregulates an E3 ligase that targets β-catenin for proteasome-dependent degradation. Previous studies suggested that CBL, BTRC, JADE1, and TRIM33 were ubiquitin E3 ligases targeting β-catenin (21–24). However, none of the genes were upregulated upon FOXF2 overexpression (Supplementary Fig. S10). To identify the E3 ligase regulated by FOXF2 and responsible for β-catenin degradation, we performed a Human Ubiquitin Ligases RT² Profiler PCR Array covering 370 ubiquitin ligase genes. With a cut-off value of 1.5-fold change, 36 genes were upregulated and 5 genes were downregulated upon FOXF2 overexpression (Fig. 5A). The PCR array result was further validated by real-time PCR and three genes (IRF2BPL, KLHL22, and ASB2) were confirmed to be upregulated at mRNA level upon FOXF2 overexpression (Fig. 5A). Although all the 3 E3 ligases suppressed TOPflash luciferase activity, only IRF2BPL, but not ASB2 or KLHL22, significantly reduced the β-catenin protein level (Fig. 5B; Supplementary Fig. S11A and S11B). FOXF2 significantly induced IRF2BPL mRNA in AGS and MKN45, whereas silence of FOXF2 in HGC27 and MG803 suppressed IRF2BPL mRNA (Fig. 5A). Ectopic expression of IRF2BPL increased endogenous β-catenin ubiquitination (Fig. 5C). Conversely, knockdown of IRF2BPL significantly reduced endogenous β-catenin ubiquitination level (Fig. 5D). Moreover, immunoprecipitation assay showed that IRF2BPL directly interact with β-catenin (Fig. 5E). To further confirm this, we performed a rescue assay by using two siRNAs targeting IRF2BPL in two gastric cancer cell lines to rule out the off-target effect and cell-specific effect. Overexpression of FOXF2 significantly downregulated β-catenin protein and suppressed its activation, whereas depletion of IRF2BPL abrogated the effect, suggesting that IRF2BPL is an important effector downstream of FOXF2 in regulating of β-catenin (Fig. 5F). IRF2BPL overexpression suppressed the activity of Wnt signaling as demonstrated by TOPflash luciferase reporter (Fig. 5G). Moreover, IRF2BPL significantly repressed the LEF1 promoter luciferase reporter activity and c-myc mRNA expression (Fig. 5G). Taken together, our data revealed that FOXF2 induces E3 ligase IRF2BPL transcription, which in turn increasing β-catenin ubiquitination and degradation.

We then examined whether FOXF2 directly induced IRF2BPL transcription. FOXF2 binding motif with a core sequence “AAACA” was identified using transcription factor motif prediction analysis jaspar (http://jaspar.genereg.net; Fig. 5H). We then performed a chromatin immunoprecipitation-PCR assay (ChIP-PCR) by using 10 sets of primers coving the promoter region of IRF2BPL form −3000 to +1500 (Fig. 5H, left). ChIP-PCR demonstrated that FOXF2 could bind to the IRF2BPL promoter region at least four sites (Fig. 5H, right). Meanwhile, we also performed ChIP-PCR analysis and found that FOXF2 could bind to the promoter region of KLHL22 and ASB2, suggesting that FOXF2 transcriptional activates KLHL22 and ASB2 (Supplementary Fig. S12; Supplementary Table S1). Furthermore, we cloned the promoter region (−2700 to TSS) into pGL3-basic plasmid and performed a luciferase activity assay. Wild-type FOXF2 but not the mutant ΔFOXF2 significantly activated the luciferase reporter, suggesting that FOXF2 activated IRF2BPL transcription (Fig. 5I). This was further confirmed by using the pGL3-basic plasmid harboring the 12-bp core “AAACA” motifs (Supplementary Fig. S13). H3K27Ac has been identified as an active regulatory histone modification marker. To further confirm FOXF2 binds and activates IRF2BPL transcription, we carried out H3K27Ac ChIP-qPCR to evaluate IRF2BPL promoter activity. FOXF2 significantly increased the level of H3K27Ac in the promoter of IRF2BPL, suggesting that FOXF2 positively regulated IRF2BPL gene transcription (Fig. 5J, left). Western blot analysis also showed that ectopic expression of FOXF2 increased IRF2BPL protein level (Fig. 5J, right). In addition, IRF2BPL mRNA expression was significantly downregulated in human gastric cancer tissues compared with the adjacent normal tissues (Fig. 5K). IRF2BPL mRNA showed a positive correlation with FOXF2 in gastric cancer in our Chinese cohort (Spearman ρ = 0.42, P < 0.05) and TCGA cohort (Spearman ρ = 0.38, P < 0.001; Fig. 5K). These findings collectively suggested that FOXF2 directly upregulates E3 ligase IRF2BPL transcription for β-catenin ubiquitination degradation (Fig. 5L).

We evaluated FOXF2 methylation in 103 primary gastric cancer tissues by BGS. Methylated FOXF2 was detected in 50.5% of primary GCs (52 of 103). We next determined the association between FOXF2 methylation and clinicopathologic features such as age, gender, differentiation, lymph node metastasis, and tumor–node–metastasis (TNM) stage, but no correlations were found (Supplementary Table S4). However, FOXF2 methylation was associated with an increased risk of cancer-related death by univariate Cox regression (RR, 2.47 (95% CI, 1.39–4.47), P = 0.012; Table 1). As expected, the TNM stage was also a significant prognostic factor (P < 0.01). In particular, after adjustment for potential confounding factors, FOXF2 methylation was found to be an independent risk factor for shortened survival in patients with gastric cancer by multivariate Cox regression analysis [RR, 1.90 (95% CI, 1.05–3.48), P = 0.035; Table 1)]. Kaplan–Meier survival curves showed that gastric cancer patients with FOXF2 methylation had significantly poorer overall survival than patients without methylation based on the log-rank test (P < 0.05). After stratification by TNM stage, FOXF2 methylated patients had significantly shorter survival in stages I–II (P < 0.05) but not in stages III–IV (P < 0.05; Fig. 6A). TCGA cohort of 243 recurrence-free gastric cancre patients verified the prognostic significance of FOXF2. FOXF2 promoter methylation was also an independent predictor of poor survival at the early stage of gastric cancer patients from TCGA cohort (based on average value of probe cg06005891, cg04187121, cg12611423, and cg19519310; Fig. 6B; Table 1). These findings indicated that FOXF2 promoter hypermethylation predicts a poor prognosis in patients with gastric cancer, especially in the early stages.

In this study, we first demonstrated that FOXF2 was generally expressed in normal human stomach tissues, but frequently silenced in gastric cancer cell lines and primary gastric cancer tumor tissues. We showed that the silencing of FOXF2 was regulated by promoter hypermethylation. In keeping out our identification, hypermethylated CpG island of FOXF2 promoter was also reported in leukemia, breast cancer, colorectal cancer, and kidney cancer (5, 25). Demethylation treatment by DNA methyltransferase inhibitor 5-Aza successfully restored the expression of FOXF2. 5-Aza (decitabine) incorporates into DNA strands upon replication and irreversibly blocks DNA methyltransferases function. 5-Aza has been under clinical study as an anticancer treatment since 1960s. However, the effect of 5-aza alone on solid tumor is lower than conventional therapy. Recent clinical trials are conducted to test the role of 5-aza in various combinations for intensification chemotherapy (epirubicin, cisplatin, fluorouracil) and Immunotherapy (Nivolumab) in solid tumors (https://clinicaltrials.gov/). TSA is a selective inhibitor of the class I and II histone deacetylase (HDAC) families. Previous studies indicated that TSA showed an antiproliferation effect in gastric cancer cells and xenograft nude mice model (26, 27). Although TSA has not been tested in a clinical setting, vorinostat, a pan-HDAC inhibitor structurally similar to TSA, has been proved by FDA for the treatment of relapsed and refractory cutaneous T-cell lymphoma (28).

A series of in vitro and in vivo functional experiments revealed that FOXF2 possesses a tumor-suppressive function in gastric cancer. The overexpression of FOXF2 could suppress the cell growth in two gastric cancer cell lines in vitro and in nude mice tumorigenesis in vivo; while the knockdown of FOXF2 promoted the cell growth. FOXF2 suppressed gastric cancer cell growth was mediated by inhibiting G₁–S cell-cycle transition and inducing cell apoptosis (Fig. 2). FOXF2 suppressed the migration and invasion ability of gastric cancer. In keeping with our finding, previous study also indicated that FOXF2 deficiency induced the epithelial–mesenchymal transition (EMT) of basal-like breast cells (29). These results strongly suggest that FOXF2 acts as a novel tumor suppressor in gastric cancer.

Luciferase reporter assay demonstrated that FOXF2 could suppress the TOPflash activity in gastric cancer cells (Fig. 3). Formation of LEF1/β-catenin complex is a prerequisite for the transcription of Wnt target gene and its formation is mainly regulated by β-catenin protein levels (30). β-Catenin was greatly reduced by FOXF2 without alteration of β-catenin mRNA expression. These results indicated that FOXF2 reduces β-catenin protein expression through posttranscriptional mechanism. In the absence of Wnt stimulation, cytoplasmic β-catenin is constitutively phosphorylated via GSK-3β. Phosphorylated β-catenin is ubiquitinated and degraded by the proteasome (30, 31). Our results showed that proteasome inhibitor MG132 could block FOXF2-induced degradation of β-catenin and ubiquitination assay confirmed that FOXF2 caused β-catenin polyubiquitylation. Thus, FOXF2 induced β-catenin degradation through ubiquitin–proteasome pathway. Moreover, we examined whether FOXF2-induced β-catenin degradation was GSK-3β independent. Overexpression of FOXF2 increased β-catenin S33Y degradation and suppressed the activity of Wnt signaling induced by β-catenin S33Y, which is GSK-3β degradation-resistant form of β-catenin. We also validated our conclusion by using GSK-3β siRNA and inhibitor. Therefore, these findings suggested that FOXF2 interrupted the interaction between LEF1 and β-catenin in the nucleus. FOXF2 promoted β-catenin degradation via ubiquitin–proteasome pathway in GSK-3β–independent fashion.

Previous studies suggested that CBL, BTRC, JADE1, and TRIM33 were ubiquitin E3 ligases targeting β-catenin. However, none of these E3 ligases were upregulated upon FOXF2 expression, suggesting a novel mechanism contributed to β-catenin degradation. On the basis of Human Ubiquitin Ligases RT² Profiler PCR Array and subsequent validation assays, IRF2BPL was identified as the most promising E3 candidate for regulating the abundance of β-catenin. Indeed, overexpression of IRF2BPL significantly reduced β-catenin protein without alteration of its mRNA expression. Moreover IRF2BPL interacted with β-catenin and significantly increase β-catenin ubiquitination. Future researches are needed to decipher the physical interplay of IRF2BPL and β-catenin and other protein factors potentially being involved in the regulation process. As FOXF2 is a nuclear transcriptional factor, the direct relationship between FOXF2 and IRF2BPL was examined. FOXF2 binding motif was predicted and eight sites with DNA binding sequence of “AAACA” were identified in the IRF2BPL promoter region. ChIP-PCR assay confirmed FOXF2 could bind to at least four IRF2BPL promoter region (#3, #4, #6, #7). IRF2BPL promoter luciferase assay and H3K27Ac ChIP assay further indicated that FOXF2 positively regulated IRF2BPL expression. Of note, IRF2BPL may not be the only E3 ligase or pathway that mediating β-catenin inhibition. KLHL22 and ASB2 significantly suppressed TOPflash reporter activity, suggesting that at least these two E3 ligases may also played a negative role in regulating of Wnt signaling pathway.

Taken together, we revealed a novel FOXF2-IRF2BPL-β-catenin signaling axis in suppressing Wnt signaling activity in gastric cancer. FOXF2 transcriptionally binds and upregulates E3 ligase IRF2BPL, which in turn interacts with β-catenin for its ubiquitination and degradation. FOXF2 also interrupts the interaction between LEF1 and β-catenin in the nucleus. These effects of FOXF2 contribute to the inhibition of Wnt signaling activity to suppress gastric cancer cell growth (Fig. 5L).

We finally investigated the clinical importance of FOXF2 methylation in 103 gastric cancer patients and found that 50.5% of them had FOXF2 promoter hypermethylation. FOXF2 methylation was an independent risk factor of poor survival in patients with gastric cancer by multivariate Cox regression analysis (Table 1). The disease-free survival of patients with FOXF2 methylation was significantly shorter than that of other patients with gastric cancer (Fig. 6A) by Kaplan–Meier survival curve analysis. In particular, methylation was significantly associated with shorter survival for patients with stage I/II gastric cancer (Fig. 6A). The clinical outcome of gastric cancer varies greatly depending on the aggressiveness of individual tumors. Many patients experience disease recurrence following radical surgery. Although adjuvant chemotherapy may benefit patients with TNM stage I/II gastric cancer, its role remains controversial due to the lack of data showing a definite benefit in this group of patients. Thus, additional prognostic biomarkers may provide better risk assessment that can guide personalized chemotherapy. Promoter methylation has been reported as a promising predictive biomarker in gastric cancers (32–34). Our results suggest that FOXF2 hypermethylation may serve as a new prognostic marker for patients with early gastric cancer.

In conclusion, we have identified a novel tumor suppressor FOXF2, which is commonly inactivated by promoter methylation in gastric cancer. FOXF2 suppresses gastric cancer growth by inhibiting Wnt signaling through FOXF2–IRF2BPL–β-catenin axis. FOXF2 directly binds to E3 ligate IRF2BPL promoter and upregulates IRF2BPL. IRF2BPL then interacts with β-catenin for its ubiquitination and degradation. FOXF2 methylation is associated with shorter survival in gastric cancer patients and may serve as a prognostic biomarker especially for early stage of gastric cancer patients.

No potential conflicts of interest were disclosed.

Conception and design: Y. Dong, J. Yu

Development of methodology: Y. Dong, Y. Zhang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Higashimori, Y. Dong, Y. Zhang, W. Kang, J. Yu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Higashimori, Y. Dong, G. Nakatsu

Writing, review, and/or revision of the manuscript: A. Higashimori, Y. Dong, Y. Zhang, S.S.M. Ng, J. Yu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Dong, J.J.Y. Sung, J. Yu

Study supervision: T. Arakawa, J.J.Y. Sung, F.K.L. Chan, J. Yu

This project was supported by research funds from RGC GRF Hong Kong (472613, 14106415, 14111216 to J. Yu), HMRF Hong Kong (1195728 to J. Yu), 135 program project (2016YFC1303200 to F.K.L. Chan), Shenzhen Virtual University Park Support Scheme to CUHK-SZRI (to J. Yu), and National Natural Science Foundation of China (81502436 to Y.J. Dong).

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.