The GSE45670 and GSE161533 datasets were derived from the NCBI/GEO database to analyze differentially expressed lncRNAs in ESCC (https://www.ncbi.nlm.nih.gov/geo/). Prediction results are presented via Venny 2.1 (https://bioinfogp.cnb.csic.es/tools/venny/index.html). GEPIA (http://gepia.cancer-pku.cn/) was used to analyze FOXP4-AS1 expression in various cancers preliminarily. UCSC (http://genome.ucsc.edu/) was used to examine the abundance of H3K4me3 in the FOXP4 promoter. RPISeq (http://pridb.gdcb.iastate.edu/RPISeq/) was used to predict the binding ability of MLL2 with FOXP4-AS1. Animal TFDB3.0 (http://bioinfo.life.hust.edu.cn/AnimalTFDB/#!/) was used to find the motif sequence of FOXP4.

With informed consent, tissue samples were collected from ESCC patients who underwent surgery at the Fourth Hospital of Hebei Medical University from 2015 to 2016. None of the patients received local or systemic treatment before surgery. Two pathologists confirmed the samples. A portion of the sample was placed in RNA later solution and stored at -80°C. Another portion was fixed in 10% neutral formalin and made into a wax block. The Ethics Committee of the Fourth Affiliated Hospital of Hebei Medical University accepted the study. The specific clinicopathological data of the patients are shown in Table 1 .

Human ESCC cell lines (Eca109, TE1, TE13, YES-2 and Kyse150) and esophageal normal epithelial cell lines (HEEpiCs) were acquired by the American Type Culture Collection. All cells were cultured in RPMI 1640 (Gibco) medium in 10% FBS (Invitrogen), and placed in an incubator (Thermo Fisher) at 37°C with 5% CO₂.

Total RNA from tissues and cells was extracted using TRIzol reagent (Invitrogen) according to the protocol. cDNA was generated using the Transcriptor First Strand cDNA Synthesis Kit (Roche). qRT–PCR was performed on a StepOne Plus real-time quantitative PCR system using GoTaq^® qPCR Master Mix (Promega). The endogenous control of mRNA/lncRNA is ACTH. The 2^-ΔΔCT method was used to calculate RNA expression. Each sample was tested in triplicate, and the average was taken. The primer sequences used are in Table 2 .

To clarify the cellular localization of FOXP4-AS1, cytoplasmic and nuclear fractions of Kyse150 and YES-2 cells (1×10⁷) were collected using the Nuclear/Cytolysis Fractionation Kit (BioVision) under the manufacturer’s guidelines. GAPDH and U6 were used as cytoplasmic/nuclear localization controls.

The shRNAs (GenePharma) were designed to knockdown FOXP4-AS1, FOXP4 and MLL2. The sh-NC plasmid was included as a control. The overexpression plasmid was synthesized from GenScript, and the pcDNA3.1-NC plasmid was used as a control. Plasmids were transfected into cells in accordance with the Lipofectamine 2000 (Invitrogen). Transfected cells were used for qRT–PCR validation or further experiments.

The MTS and colony formation assays were applied to detect the cell proliferation capacity. For the MTS assay, transfected cells were seeded into 96-well plates at 1×10³ per well. After 0, 24, 48, 72 and 96 hours of incubation 20 μl (500 μg/mL) MTS reagent was added to each well and incubated for 2 hours in the CO₂ incubator. A Spark^® multimode microplate reader (Tecan) was used to measure the absorbance at 492 nm wavelength. For the colony formation assay, transfected cells were inoculated in six-well plates at 3×10³ cells per well and cultured for 7–14 days. Cells were fixed and stained with paraformaldehyde or 0.5% crystalline violet. A colony was a cluster of >50 cells, and the number of colonies was counted manually.

The migration assays were executed using a nonmatrix gel-coated chamber (Corning) with an 8 μm pore membrane. A total of 1×10⁵ cells were inoculated into the upper chamber, and the lower chamber was 600 µl complete culture medium. After incubation for 24 hours, cells were fixed with paraformaldehyde and stained with crystal violet solution. Three areas were randomly selected for cell counting under a microscope (Leica). The invasion assay was performed as described above except for 50 µl Matrigel (Corning) in the upper chamber.

Apoptosis assays were executed using Annexin V-FITC/PI (BD Bioscience) double-staining reagent. Transfected cells were collected 48 hours later, washed twice with PBS and adjusted to 1×10⁶ cells per ml. Subsequently, cells were stained for apoptosis by adding 10 µl Annexin V-FITC and 5 µl PI under light-protected conditions. The cell cycle assays were performed using PI (Multi Science) single staining reagent. First, cells were treated with Triton X-100 and stained for 30 min with 500 µl PI at room temperature. Apoptosis and cell cycle were assessed by a flow cytometer (BD Bioscience).

Cells were lysed in RIPA buffer (Solarbio) containing PMSF. Protein quantification was performed using the BCA Protein Quantification Kit (Solarbio). The protein sample was blended with the protein loading buffer and heated at 99°C for 5 minutes. The denatured proteins were separated by 10% SDS–PAGE and then transferred to a polyvinylidene difluoride (PVDF) membrane (Millipore). The PVDF membrane was placed in 5% skim milk for 1 hour and incubated at 4 ℃ overnight with the following primary antibodies: FOXP4 (1:1000, Abcam), GAPDH (1:10000, Abcam), MLL2 (1:1000, Abcam), E-cadherin (1:1000, Elabscience), vimentin (1:1000, Elabscience), and β-catenin (1:1000, Elabscience). Subsequently, the membranes were incubated with goat anti-rabbit IgG (KPL) for 1 hour, and then ECL reagent (Solarbio) was added for observation.

Paraffin-embedded tumor and normal tissues were sectioned and immunostained to determine FOXP4 protein expression using a rabbit anti-human FOXP4 antibody (1:1000, Abcam). Scoring criteria were evaluated according to previous reports (15). Three experienced pathologists, without knowledge of the clinical data, investigated the sections simultaneously.

Animal TFDB3.0 predicted the binding site of FOXP4 on the β-catenin promoter with two binding sites, and designed primers to amplify the fragment containing the binding site with upstream and downstream endonucleases HindIII and EcoRI. The fragments were ligated into pmirGL3-Basic and named pmirGL3-β-catenin-1 or pmirGL3-β-catenin-2. The mutant primers were devised in NEBase changer (http://nebasechanger.neb.com/) and synthesized by Sangon Biotech. The target fragments were successfully colony into pmir-GL3-Basic and named pmirGL3-β-catenin-1 (MUT) or pmirGL3-β-catenin-2 (MUT). The sequences of all constructed plasmids were sequenced in perspective.

PmirGL3-β-catenin-1/2(WT) and pmirGL3-β-catenin-1/2(MUT) were cotransfected with FOXP4 in Kyse150. A dual luciferase reporter system (Promega) was used to measure the luciferase activity of cells after 48 h.

The binding of FOXP4-AS1 to MLL2 was detected by RNA immunoprecipitation method using MLL2 antibody (Abcam) and the Magna RIP™ RNA-Binding Protein Immunoprecipitation Kit (Millipore) according to the manufacturer’s instructions. The IgG antibody was used as negative control. The purified RNA was subjected to qRT-PCR analysis.

ChIP assays were executed by the EZ-Magna ChIP kit (Millipore) and followed the manufacturer’s protocol. Cells were fixed in 4% PFA and quenched with glycine for 10 min. DNA was broken down to 200 to 600 bp using ultrasound. Immunoprecipitation of the lysate was performed with anti-H3K4me3 (Abcam), anti-MLL2 (Abcam), rabbit IgG or anti-FOXP4 (Abcam).

The experimental data for each one were performed individually and repeated three times. The results were statistically analyzed using SPSS 21.0 (IBM) and GraphPad Prism 8 (GraphPad), and shown as the mean ± SD. Student’s t-test and one-way ANOVA assessed significant differences between groups. Pearson correlation analysis was used to analyze the correlation between the FOXP4-AS1, FOXP4 and β-catenin. Survival curves were measured by the Kaplan–Meier method. Two-tailed values of P<0.05 were considered statistically significant..

The differential lncRNAs in GSE45670 were screened by log|FC|>1 and P<0.05. A total of 457 differentially expressed lncRNAs were found, including 242 downregulated genes and 215 upregulated genes, among which the log|FC| value of FOXP4-AS1 was 2.08 ( Figure 1A ). We also found that FOXP4-AS1 was highly expressed in GSE161533 and GSE45670 ( Figure 1B ). The DEPIA revealed that FOXP4-AS1 was upregulated in most malignancies ( Figure 1C ). The qRT–PCR results indicated that FOXP4-AS1 was overexpressed in ESCC tissues ( Figure 1D ), and correlated with lymph node metastasis and TNM stage ( Figure 1E ). FOXP4-AS1 in six ESCC cell lines (Kyse150, YES-2, Kyse170, TE13, Eca109 and TE1) was considerably higher than the HEEpiC cells ( Figure 1F ). The expression of FOXP4-AS1 was highest in YES-2 cells, so YES-2 cells were used for knockdown experiments, and Kyse150 cells were used for overexpression experiments. Receiver operating characteristic (ROC) curve analysis showed that FOXP4-AS1 had a high-value area under the curve (AUC) of 0.8679 (0.8060 to 0.9298), a sensitivity of 85.50% and a specificity of 84.10% ( Figure 1G ) in distinguishing ESCC from normal tissues. Application of the Kaplan–Meier method revealed a significant reduction in overall survival (OS) ( Figure 1H ) and recurrence-free survival (RFS) ( Figure 1I ) in 69 ESCC patients with high expression of FOXP4-AS1. Collectively, these results suggest that FOXP4-AS1 may play a role as an oncogene in ESCC development, and may be a biomarker for ESCC diagnosis, treatment and prognosis.

To evaluate the biological function of FOXP4-AS1 in ESCC. Knockdown of FOXP4-AS1 in YES-2 using shRNA, we chose sh-FOXP4-AS1–2 for follow-up experiments because it exhibited relatively strong knockdown efficiency ( Figure 2A ). Transfection of pcDNA3.1-FOXP4-AS1 into Kyse150 greatly increased FOXP4-AS1 compared to the empty vector control (pcDNA3.1-NC) ( Figure 2B ). MTS and colony formation assays revealed that the proliferation ability of YES-2 cells transfected with sh-FOXP4-AS1 was markedly inhibited ( Figures 2C, E ). The proliferation ability of Kyse150 cells transfected with pcDNA3.1-FOXP4-AS1 was enhanced ( Figures 2D, F ). Consistent with the migration and invasion assays, fewer trans-well cells were discovered in the sh-FOXP4-AS1 group than the sh-NC group ( Figures 2G, I ). At the same time, transfection of pcDNA3.1-FOXP4-AS1 in Kyse150 had the opposite effect ( Figures 2H, J ). Flow cytometry was used to detect the effect of FOXP4-AS1 on the ESCC cell cycle and apoptosis. The data showed that the early apoptotic cells in the sh-FOXP4-AS1 group grew in number ( Figure 2K ). The cell number of G0/G1 phase was augmented, and S+G2/M phase was reduced in YES-2 cells transfected with sh-FOXP4-AS1 ( Figure 2L ).

FOXP4 mRNA and protein were dramatically highly expressed in ESCC, and correlated with lymph node metastasis and TNM stage ( Figures 3A–C ). Knockdown of FOXP4-AS1 in YES-2 drastically reduced FOXP4 mRNA and protein. Overexpression of FOXP4-AS1 at Kyse150 promoted an increase in FOXP4 ( Figures 3D, E ; S1A ). Pearson correlation analysis revealed that FOXP4-AS1 positively correlated with FOXP4 mRNA in ESCC ( Figure 3F ). To further investigate the possible molecular mechanisms of FOXP4-AS1 involvement in cancer development, we examined its subcellular localization. FOXP4-AS1 was mainly located in the nucleus of ESCC cells ( Figures 3G, H ).

Numerous studies have shown that antisense lncRNAs can cis-regulate sense mRNAs’ expression and molecular mechanism (16–18). Overexpression of FOXP4-AS1 in Kyse150 cells upregulated the mRNA and protein of FOXP4. We considered whether FOXP4-AS1, located in the nucleus, might regulate FOXP4 by binding to some nuclear factors, such as transcription factors or epigenetic regulatory enzymes. We observed that the promoter of FOXP4 was enriched in H3K4me3 signaling, implying that FOXP4 might be affected by histone methylation ( Figure 4A ). Myeloid/lymphoid or mixed-lineage leukemia (MLL) family genes are essential enzymes for modifying H3K4 methylation, so we speculate that FOXP4-AS1 may regulate FOXP4 by binding to MLL family genes. The binding capacity of MLL1-MLL5 to FOXP4-AS1 was predicted by RPISeq, showing that MLL2 binds optimally to FOXP4-AS1 (RF:0.70, SVM:0.95). RIP assays confirmed that FOXP4-AS1 could bind to the MLL2 protein ( Figure 4B ). MLL2 was highly expressed in ESCC ( Figure 4C ), and positively correlated with FOXP4-AS1 in ESCC by Pearson analysis ( Figure 4D ). Moreover, MLL2 overexpression increased the mRNA and protein of FOXP4. MLL2 and FOXP4-AS1 cotransfection increased FOXP4 at the transcriptional and translational levels, indicating that MLL2 and FOXP4-AS1 have synergistic effects on FOXP4 regulation ( Figures 4E–H ; S1B, C ). In addition, FOXP4 was positively associated with MLL2 in ESCC ( Figure 4I ). ChIP assay data further demonstrated that upregulated FOXP4-AS1 promoted the enrichment of MLL2 and H3K4me3 in the FOXP4 promoter ( Figure 4J ). However, MLL2 did not affect the expression of FOXP4-AS1 ( Figures S2A, B ). Thus, FOXP4-AS1 enriched MLL2 and H3K4me3 in the FOXP4 promoter, which induced transcription of FOXP4.

The effect of FOXP4 on the biological function of ESCC was examined by knocking down FOXP4 with shRNAs in YES-2 cells. We chose sh-FOXP4–3 for subsequent experiments, due to its relatively more robust knockdown efficiency ( Figures 5A, B ; S1D ). Transfection of pcDNA3.1-FOXP4 in Kyse150 resulted in substantially higher mRNA and protein levels of FOXP4 than pcDNA3.1-NC ( Figures 5C–E ; S1E ). The proliferation ability of YES-2 cells transfected with sh-FOXP4 was greatly inhibited as shown by MTS and colony formation assays ( Figures 5F, H ). Transfection with pcDNA3.1-FOXP4 substantially enhanced the proliferation ability of Kyse150 cells ( Figures 5G, I ). Migration and invasion assays had consistent results, with fewer trans-well cells in the sh-FOXP4 group than the sh-NC group ( Figures 5J, L ). While transfection of pcDNA3.1-FOXP4 had the opposite effect ( Figures 5K, M ).

EMT plays a critical role in the invasion and distant metastasis of malignant tumors, including ESCC (19–21). We aimed to clarify the function of FOXP4 in mediating EMT in ESCC. qRT–PCR and WB showed that the expression of vimentin and β-catenin was considerably lower in the sh-FOXP4 group than the control group, and overexpression of FOXP4 increased their expression, while E-cadherin had the opposite effect ( Figures 6A–D ; S1F–H ). We found that β-catenin was upregulated in ESCC ( Figure 6E ), and correlated with lymph node metastasis and TNM stage ( Figure 6F ). Pearson analysis revealed that FOXP4 is related to β-catenin in ESCC ( Figure 6G ). Animal TFDB3.0 predicted that FOXP4 could act as a transcription factor of β-catenin, and had two binding sites ( Figure 6H ). The luciferase reporter assay demonstrated that FOXP4 increased the luciferase activity of pmirGL3-β-catenin-1/2 (WT) in Kyse150 cells, whereas the mutant did not ( Figure 6I ). ChIP results showed that the expression of PCR-amplified products was substantially higher in the FOXP4 antibody group than the control group, suggesting that FOXP4 had binding site activity with the β-catenin promoter ( Figure 6J ). However, the promoter region of FOXP4-AS1 did not have a binding site for FOXP4, and the experimental results indicated that FOXP4 did not affect FOXP4-AS1 ( Figures S2C, D ). Therefore, we inferred that FOXP4 promoted β-catenin transcription by binding to the β-catenin promoter.

We found that FOXP4 acts as a transcription factor for β-catenin to verify further whether FOXP4-AS1 regulates β-catenin expression and ESCC biological function through the MLL2/FOXP4 axis. FOXP4-AS1/FOXP4 knockdown and overexpression plasmids were constructed and cotransfected into Kyse150 cells. It was revealed that FOXP4-AS1 promoted β-catenin expression, and the highest expression of β-catenin was observed cotransfection with FOXP4-AS1/FOXP4. However, sh-FOXP4-AS1/FOXP4 eliminated this effect ( Figure 7A ; S1I ). Cell function experiments showed that cotransfection of FOXP4-AS1/FOXP4 increased the proliferation, migration and invasion ability of ESCC cells. Conversely, sh-FOXP4-AS1/FOXP4 partially rescued this effect ( Figures 7B–E ). These results suggest that FOXP4-AS1 regulates β-catenin expression and biological functions through the MLL2/FOXP4 axis ( Figure 7F ).