RNA m6A demethylase FTO-mediated epigenetic up-regulation of LINC00022 promotes tumorigenesis in esophageal squamous cell carcinoma

LncRNA expression datasets from two independent ESCC cohorts, GSE75241↗ and TCGA-ESCC used as discovery datasets, were retrieved from the Gene Expression Omnibus (GEO) and The Cancer Genome Atlas (TCGA) databases, respectively. GSE75241↗ contains 15 pairs of ESCC tumor tissues and normal tissues. TCGA-ESCC has 81 ESCC tumor samples and 11 normal samples. Differentially expressed LncRNAs (DELncRNAs) profiles of GSE75241↗ and TCGA-ESCC were analyzed by using Limma [32] and DESeq2 [33], respectively. LINC00022 expression data in another two ESCC cohorts obtained from the public databases GEPIA [34] and CRN (http://syslab4.nchu.edu.tw↗) was used as validation datasets.

The ESCC cohort in this study included forty-four pairs of surgically resected ESCC tissue specimens from Linzhou Cancer Hospital, Henan province of China were collected and performed to detect the differential gene expression. Thirty-six pairs of ESCC specimens with relatively complete clinical data were utilized to analyze the clinical relevance of gene expression. All patients signed informed consent and did not receive any chemotherapy or radiotherapy before surgery. The protocols were approved by the Ethical Review Committee of Zhengzhou University.

The diagnostic value of LINC00022 in ESCC was analyzed by receiver operating character (ROC) curve based on the expression level of LINC00022 in tumor tissues and normal tissues in each ESCC cohort. The sensitivity is represented by the area under the curve (AUC), and the higher the AUC, the greater the diagnostic value.

The secondary structure of the full-length sequence of LINC00022 transcript at minimum free energy (MFE) was computerized by the RNAfold web server (http://rna.tbi.univie.ac.at//cgi-bin/RNAWebSuite/RNAfold.cgi↗).

The potential RNA-protein interactions between LINC00022 and p16, p21, p53 were predicted by the machine learning classifier RPISeq [35] based on Random Forest (RF) or Support Vector Machine (SVM) classifiers.

The putative biological function of LINC00022 in ESCC was analyzed by gene enrichment method. We first obtained a thousand LINC00022-associated genes in ESCC tissues from GEPIA database, and then used WebGestalt (http://www.webgestalt.org/↗) to analyze the clustering properties of these genes based on the KEGG pathway. The result of pathway enrichment analysis is shown as a bubble diagram.

The genome copy number variation (CNV) data of LINC00022 was obtained from TCGA and analyzed by GISTIC2.0 [36]. The correlation between LINC00022 expression and CNV level of LINC00022 in 159 cases of esophageal cancer tissues was analyzed by Person’s Coefficient. DNA methylation data of LINC00022 promoter was also retrieved from the TCGA project and analyzed by Limma package. The correlation between LINC00022 expression and methylation level of LINC00022 promoter in 161 cases of esophageal cancer tissues was then analyzed.

Human ESCC cell lines KYSE150 and TE1 were purchased from Procell (Wuhan, China). KYSE70, KYSE450 and one human esophageal epithelial cell line Het-1A were kindly provided by the Bioengineering and Transformation Laboratory of Zhengzhou University. All cell lines were identified by short tandem repeat (STR) analysis and cultured in RPMI 1640 (Procell, China) basic medium supplemented with 1% penicillin-streptomycin solution (Procell, China) and 10% fetal bovine serum (FBS) (LONSERA, Uruguay) at 37 °C containing 5% CO₂.

The recombinant plasmids for over-expressing YTHDF2 (OE-YTHDF2) and the negative control plasmids (OE-NC) were purchased from GeneChem (Shanghai, China). Three small interfering RNAs (siRNAs) against LINC00022 (si-022#1, si-022#2, si-022#3) and the negative control siRNAs (si-NC) were purchased from Public Protein/Plasmid Library (PPL) (Nanjing, China). The siRNAs targeting YTHDF2 (si-YTHDF2) were synthesized by GenePharma (Shanghai, China). Lipofectamine 3000 (Invitrogen, USA) and INTERFERin (Polyplus, France) were used for transfecting plasmids and siRNAs into cells, respectively.

Recombinant lentiviruses for knocking down LINC00022 or FTO (sh-LINC00022 or sh-FTO) and lentiviruses for over-expressing LINC00022 or FTO (OE-LINC00022 or OE-FTO) as well as the corresponding control viruses were purchased from GeneChem (Shanghai, China). HiTransG A or HiTransG P solution (GeneChem, China) was used to increase the efficiency of cell infection.

Total RNAs from cells or tissues were extracted by Trizol (Invitrogen, USA) reagent and quantified by Nanodrop 2000 (Thermo, USA). After cDNA synthesis with the NovoScript All-in-one SuperMix (Novoprotein, China), the 7500 Fast Real-Time System (AB, USA) and NovoStart SYBR SuperMix (Novoprotein, China) were employed to perform the qRT-PCR assay. The house-keeping gene GAPDH was used as an internal reference. The qRT-PCR primer sequences used in this study are shown in Supplementary Table. 1

Cell proliferation was evaluated by the CCK-8, EdU staining and colony formation assays. For the CCK-8 assay, cells were inoculated in 96-well plates and maintained in an incubator for a specified period of time. 10 μL of CCK-8 solution (7sea biotech, China) was added to each well 3 h prior to measurement at 450 nm. For the EdU staining assay, cells were labeled with EdU staining solution (Abbkine, USA) at 37 °C for 4 h, fixed in 4% formaldehyde and permeabilized with 0.5% Triton X-100. Then the cells were incubated in Click-iT mixture and re-stained with DAPI solution, and finally detected under a fluorescent microscope. For the colony formation assay, cells were seeded in 12-well plates and maintained in an incubator for 8 to 12 days to form the colonies. After fixation and crystal violet staining, the colonies in each well were photographed and then counted by the Image J software.

The Annexin V-FITC/PI double staining kit (7sea biotech, China) and PI staining kit (7sea biotech, China) were used for testing cell apoptosis and cell cycle, respectively. Cells stained with Annexin V-FITC and PI solution in turn were subjected to evaluate apoptotic rates by a flow cytometer (Beckman, USA). Cells fixed in 70% ethanol and incubated with PI solution containing RNase were used to analyze cell-cycle changes.

Total proteins were isolated from cells or tissues using the RIPA lysis buffer (Beyotime, China) and subjected to 10% polyacrylamide gel electrophoresis. After transferring to PVDF membrane, blocking with 5% skim milk, incubation with primary antibodies and secondary antibodies, the proteins on the membrane were finally detected by ECL kit (EpiZyme, China) and imaging analysis system (BioRad, USA). The primary antibodies used in this study were as follows: anti-GAPDH (Bioworld, 1:10000), anti-CDK2 (Bioworld, 1:1000), anti-Cyclin E1 (Bioworld, 1:1000), anti-CDK4 (Bioworld, 1:1000), anti-Cyclin D1 (Bioworld, 1:1000), anti-p53 (Bioss, 1:500), anti-p21 (Bioworld, 1:500), anti-p16 (Bioss, 1:1000), anti-ubiquitin (Santa, 1:500), anti-FTO (Bioss, 1:1000), anti-YTHDF2 (Abcam, 1:1000).

The transfected cells were inoculated in 6-well plates and maintained in an incubator overnight at 37 °C. Then the cells were treated with cycloheximide (CHX) (Sigma, USA) at a concentration of 100 μg/mL for the indicated times. Total proteins in each group were extracted and subjected to Western blot analysis.

The Magna RIP Kit (Millipore, USA) combined with qRT-PCR were utilized to analyze the direct interactions between LINC00022 transcript and indicated proteins. The procedures were carried out according to users’ instructions of the kit. Cells were broken up with RIP lysis buffer containing protease inhibitor cocktail and RNase inhibitors. Then the cell lysates and magnetic beads conjugated with specific antibodies or IgG were incubated in RIP immunoprecipitation buffer at 4 °C for 3 h with rotation. After digestion with protease K, the immunoprecipitated RNA samples were purified by phenol/chloroform/isoamyl alcohol and 100% ethanol, and finally detected by qRT-PCR and 1.5% agarose gel electrophoresis.

The Pierce Classic Magnetic Co-IP Kit (Thermo, USA) combined with Western blot was utilized to detect the effect of LINC00022 over-expression on ubiquitination of p21 in ESCC cells. In brief, cells were broken up with ice-cold IP lysis buffer to obtain the supernatant containing total proteins. The cell lysates were incubated with antibodies in a tube overnight at 4 °C to form the immunoprecipitation complex. Then the pre-treated protein A/G beads were added to each tube and incubated at room temperature for 1 h. Finally, the supernatant containing the target antigen collected by the magnetic stand was used for Western blot analysis.

Two reliable online tools RMBase v2.0 [37] and SRAMP [38], based on epitranscriptome sequencing information and machine learning pattern, were used to predict potential m6A modification sites on the LINC00022 transcript. The m6A2Target (http://m6a2target.canceromics.org↗) database was employed to predict m6A modification enzymes that may perturb LINC00022 expression based on MeRIP-seq or RNA-seq profiles.

The overall RNA m6A level was measured by the EpiQuik m6A RNA Methylation Kit (Colorimetric) (Epigentek, USA). In brief, total RNAs were isolated from cells or tissues and quantified by Nanodrop 2000 (Thermo, USA). Total RNA of 200 ng and binding solution were added to each well and incubated at 37 °C for 1.5 h. Then the capture antibody, detection antibody, enhancer solution, developer solution and stop solution were added to each well in turn and reacted for the indicated time. Finally, the relative m6A level in each group was compared by detecting the absorbance value at 450 nm via a microplate reader (Molecular Devices, USA).

The Magna MeRIP m6A Kit (Millipore, USA) was utilized to determine the enrichment of m6A at specific sites on the LINC00022 transcript. The fragmentation buffer was first used to fragment 300 μg of total RNA (at a concentration of 1 μg per μL) in each sample. Then the fragmented RNA and magnetic beads conjugated with m6A or IgG antibodies were incubated in the MeRIP reaction mixture at 4 °C for 2 h with rotation. Finally, the co-precipitated RNA samples containing m6A modification sites were used for qRT-PCR detection. Three specific primers were designed for MeRIP-qRT-PCR according to the 3 m6A modification sites on LINC00022 transcript with higher confidence predicted by RMBase v2.0 and SRAMP (Supplementary Fig.A-B). The primer sequences are listed in Supplementary Table. 10 1

The transfected cells were seeded in 12-well plates and maintained in an incubator overnight at 37 °C. Then the cells were treated with Actinomycin D (ActD) (Sigma, USA) at a concentration of 5 μg/mL for the indicated times. Total RNAs in each group were extracted and subjected to qRT-PCR analysis.

The 6-week-old male BALB/c nude mice were purchased from Beijing Vital River Co., Ltd. (Beijing, China). All the mice were housed in automatic air cages with free access to food and water. For in vivo tumorigenesis experiments, ESCC cells with stable over-expression or knockdown of the target gene were inoculated subcutaneously on the right dorsal side of mice. For Figs. 3A, C, 8H and I, the number of cells inoculated in each mouse was 4 × 10⁶, 1 × 10⁶, 2 × 10⁶ and 1 × 10⁶, respectively. Tumor size was measured using the length × width² × 0.5 formula. At the end of each experiment, the tumors were isolated and weighed. Each tumor was divided into three parts, one for paraffin sectioning and staining (e.g., HE or Ki-67 IHC), one for total RNA extraction, and one for total protein extraction.

All experimental data are displayed as mean ± SD from at least three replicates. SPSS 19.0 and Graphpad 7.0 software were used for statistical analysis and plotting of the data. Student’s t-test was used to compare the differences between the two groups of data. Pearson’s Coefficient was utilized to analyze the expression correlation between two genes. Survival curve and hazardous ratio (HR) for the indicated gene and ESCC cohort were detected by the Kaplan-Meier method and log-rank test. A p-value < 0.05 was considered as statistically significant.

To identify potential oncogenic LncRNAs in ESCC, we analyzed the differential expression profiles of LncRNAs (DELncRNAs) from two independent ESCC cohorts, GSE75241↗ and TCGA-ESCC. A total of 33 and 3134 DELncRNAs were analyzed from datasets GSE75241↗ and TCGA-ESCC, respectively. We compared 33 DELncRNAs from GSE75241↗ with the top-200 up-regulated DELncRNAs from TCGA-ESCC, and found that LINC00022 was the only LncRNA that was significantly elevated in both datasets (Fig. 1A). LINC00022 expression in ESCC tumor samples and normal samples in datasets GSE75241↗ (p = 5.35E-4) and TCGA-ESCC (p = 1.14E-9) was shown as box plots (Fig. 1A). These results implicated that LINC00022 plays an important role in ESCC. We then analyzed the expression of LINC00022 in our in-house cohort containing 44 pairs of ESCC tumor specimens and matched normal specimens by qRT-PCR. In line with the above dataset analysis, LINC00022 was indeed elevated in our cohort ESCC tumors (Fig. 1B-a, p = 0.0079). We also analyzed the paired results of LINC00022 expression in these 44 cases of ESCC. As shown in Supplementary Fig. 1A, LINC00022 was up-regulated in 33 ESCC tumors (75%) and down-regulated in 11 ESCC tumors (25%). When interrogating two additional ESCC datasets from GEPIA (Fig. 1B-b, p = 1.06E-43) and CRN (Fig. 1B-c, p = 6.41E-9) databases, we found that LINC00022 was consistently increased in tumor tissues compared to normal tissues. In addition, LINC00022 was likewise significantly up-regulated in tumor samples from the TCGA-EAC and TCGA-ESCA datasets (Supplementary Fig. 1B-C), implying that it also plays a role in EAC.

To gain a better understanding of the potential role of LINC00022 in human pan-cancer, we used the comprehensive database GEPIA to analyze the expression landscape of LINC00022 in 28 types of cancer other than ESCC. LINC00022 was found highly expressed in tumor tissues of CESC (p = 1.16E-6), DLBC (p = 2.83E-21), GBM (p = 2.31E-35), SKCM (p = 7.77E-89), STAD (p = 2.74E-62), and THYM (p = 6.09E-55), while low expressed in tumor tissues of TGCT (p = 4.15E-45), when compared with the corresponding normal tissues (Supplementary Fig. 2A). No significant difference was found as to the expression of LINC00022 between tumor samples and normal samples in ACC, BLCA, BRCA, CHOL, COAD, HNSC, KICH, KIRC, KIRP, LAML, LGG, LIHC, LUAD, LUSC, OV, PAAD, PRAD, READ, THCA, UCEC, and UCS (Supplementary Fig. 2B-D). Kaplan-Meier analysis from GEPIA showed that LINC00022 expression was not associated with OS in patients with CESC, DLBC, GBM, SKCM, STAD, THYM, and TGCT (Supplementary Fig. 3). This indicates that LINC00022 plays a specific role in ESCC, but not in other types of cancer.

We next analyzed the correlation between LINC00022 and clinical parameters using our 36 patients with complete information in our in-house cohort. We found that LINC00022 was up-regulated in tumor tissues of female patients or advanced stage patients, when compared with male or early stage patients (Supplementary Fig. 4A), respectively. In addition, no significant correlation was found between the expression of LINC00022 and age, differentiation, grade, lymph node metastasis or depth of invasion (Supplementary Fig. 4B-C). ROC curve was conducted to evaluate the predictive values of LINC00022 in a variety of ESCC cohorts. The AUC of LINC00022 in our cohort, GSE75241↗, TCGA-ESCC, and CRN was 0.668 (Fig. 1C, p = 0.0065), 0.8222 (Fig. 1D, p = 0.003), 0.9001 (Supplementary Fig. 5A, p < 0.0001), and 0.9795 (Supplementary Fig. 5B, p < 0.0001), respectively. LINC00022 was also showed high diagnostic values in datasets TCGA-EAC (Supplementary Fig. 5A, AUC = 0.9398, p < 0.0001) and TCGA-ESCA (Supplementary Fig. 5A, AUC = 0.9198, p < 0.0001). Kaplan-Meier analysis of ESCC cohort from GEPIA revealed that high expression of LINC00022 was significantly associated with poor patient OS (Fig. 1E, HR = 2.3, p = 0.014) and was a risk factor for unfavorable DFS (Supplementary Fig. 5C, HR = 1.9, p = 0.063). Additionally, survival analysis from TCGA cohorts depicted that patients with high LINC00022 expression had a shorter median survival than those with low LINC00022 expression in both ESCC (Supplementary Fig. 5D, HR = 1.99, p = 0.17, 25.4 months vs. 42.1 months) and EAC (Supplementary Fig. 5E, HR = 2.35, p = 0.012, 16.5 months vs. 46.7 months). Furthermore, LINC00222 was frequently up-regulated in ESCC cell lines as compared to Het-1A (Fig. 1F). Collectively, these data suggest that LINC00022 is highly expressed in ESCC, and is of diagnostic and predictive value.

To determine the oncogenic properties of LINC00022 in ESCC, we first performed gene enrichment analysis. To this end, one thousand LINC00022-associated genes were obtained from 182 ESCC sequenced samples in GEPIA database and then subjected to KEGG analysis. The LINC00022-related genes were found to be significantly enriched in biological processes such as DNA replication and cell cycle (Supplementary Fig. 6), suggesting that LINC00022 may play a crucial role in ESCC cell proliferation. We used siRNAs to specifically reduce the intracellular RNA level of LINC00022 (Fig. 2A) and the Lentivirus-mediated system to activate the expression of LINC00022 (Fig. 2B) in ESCC cells. Then, the proliferation ability of ESCC cells following LINC00022 manipulation was evaluated by CCK-8, colony formation and EdU incorporation assays. CCK-8 viability assay demonstrated that silencing of LINC00022 effectively repressed the cell survival of KYSE150 and TE1 (Fig. 2C-D), while over-expression of LINC00022 enhanced the viability of KYSE70 cells (Fig. 2E). In addition, depletion of LINC00022 significantly impaired the colony-forming capacity of ESCC cells (Fig. 2F), and enforced expression of LINC00022 elicited the opposite biological effect (Fig. 2G). Furthermore, EdU staining assay showed that LINC00022 deletion in both KYSE150 and TE1 cells suppressed their growth (Fig. 2H), whereas ectopic expression of LINC00022 markedly promoted the proliferative ability of KYSE70 cells (Fig. 2I). Together, both loss-of-function and gain-of-function experiments illustrate that LINC00022 facilitates cell proliferation of ESCC in vitro.

Given that LINC00022 is critical for sustaining the proliferation of ESCC cells, we further used the xenograft nude mice model to verify the in vivo role of LINC00022 in ESCC tumorigenesis. KYSE150 cells with Lentivirus-mediated stable knockdown of LINC00022 (sh-022) and KYSE70 cells with stable over-expression of LINC00022 (OE-022) were inoculated subcutaneously into nude mice to investigate the role of LINC00022 on the tumorigenicity of ESCC cells. Notably, LINC00022 knockdown decreased the growth capacity of KYSE150 cells in nude mice, showing smaller tumor volume and lower tumor weight (Fig. 3A). Decreased LINC00022 expression was observed in tumors of sh-022 groups (Fig. 3B). Likewise, over-expression of LINC00022 significantly increased the tumorigenicity of KYSE70 cells in nude mice, as indicated by larger tumors (Fig. 3C) given the elevated expression of LINC00022 in tumors of OE-022 groups detected by qRT-PCR (Fig. 3D). Consistent with these observations, IHC results showed that Ki-67, a molecular marker of cell proliferation, was decreased in LINC00022-deficient tumors (Fig. 3E) and up-regulated in LINC00022-augmented tumors (Fig. 3F) as compared with the corresponding controls. All these data imply that LINC00022 as a driving factor in ESCC tumorigenesis.

Accelerated cell-cycle progression is a key hallmark of malignant proliferation of cancer cells [39, 40], and targeted inhibition of cell cycle is widely considered to be an effective clinical anti-tumor strategy [41, 42]. Flow cytometry analysis revealed that siRNA-mediated silencing of LINC00022 induced changes in the cell-cycle distribution of KYSE150 and TE1 cells, resulting in an increased proportion of G0/G1 cells and a decreased ratio of G2/M cells (Fig. 4A). While, the over-expression of LINC00022 led to a decreased G0/G1 phase and prolonged G2/M phase in KYSE70 cells (Fig. 4B). Importantly, LINC00022 knockdown was not significantly associated with apoptosis in both KYSE150 and TE1 cells, as detected by Annexin V-FITC/PI double staining and PI-labeling assays (Supplementary Fig. 7A-B). Additionally, the protein expression of a panel of cell-cycle regulators in ESCC cells following LINC00022 dysregulation was examined by Western blot. Ablation of LINC00022 resulted in consistent up-regulation of CDK inhibitors p16, p21, and p53 in KYSE150 and TE1 cells (Fig. 4C-D), while over-expression of LINC00022 decreased the protein levels of p16, p21 and p53 in KYSE70 cells (Fig. 4E). Concomitantly, the protein levels of CDK2 and Cyclin E1, but not CDK4 and Cyclin D1, in ESCC cells decreased with LINC00022 deletion and increased with LINC00022 over-expression (Fig. 4C-E). These results suggest that LINC00022 regulates cell-cycle progression in ESCC cells.

Since LINC00022 manipulation affects the protein levels of CDK inhibitors in ESCC cells, we next explored its potential mechanism. Results of qRT-PCR indicated that deletion or over-expression of LINC00022 did not significantly affect the mRNA levels of p16, p21 and p53 in ESCC cells (Fig. 5A-C). The secondary structure analysis based on the minimum free energy (MFE) algorithm of the RNAfold server showed that LINC00022 contained multiple branching structures (Fig. 5D), suggesting that LINC00022 has the protein binding potential. The machine learning classifier RPISeq based on Random Forest (RF) and Support Vector Machine (SVM) classifiers predicted that the LINC00022 transcript had a high affinity with p16, p21, and p53 proteins (Fig. 5E). Binding with functional proteins is a common molecular regulatory mechanism by which LncRNAs exert their biological functions [23, 24]. Subsequent RIP-qRT-PCR assays demonstrated that LINC00022 bound directly with p21 protein, but not with p16 or p53 proteins, in ESCC cells (Fig. 5F). We next performed Western blot to detect p21 protein levels in 21 cases of ESCC (showing 14 cases) and found that p21 was reduced in most tumor samples relative to normal samples (Fig. 5G). In addition, the inverse correlation between LINC00022 transcription and p21 protein in 21 cases of ESCC tumors was analyzed by Person’s Coefficient (Fig. 5H, R = − 0.314), which was consistent with the negative regulation of p21 protein levels by LINC00022 in ESCC cells (Fig. 4C-E). Clinically, Kaplan-Meier analysis of TCGA-ESCC dataset showed that patients with high level of p21 had longer DFS (Supplementary Fig. 8A).

KEGG-based gene enrichment analysis in Supplementary Fig. 6 also showed that LINC00022 may be involved in proteasome and ubiquitin-mediated proteolysis pathways in ESCC. Therefore, we hypothesized that LINC00022 negatively regulates the protein level of p21 through the ubiquitin-proteasome pathway (UPP). Protein stability assay depicted that ectopic expression of LINC00022 shortened the half-life of p21 protein in the presence of CHX (Fig. 5I), an inhibitor of protein synthesis. The decrease in p21 protein levels caused by over-expression of LINC00022 was partially abrogated by the proteasome inhibitor MG132 (Fig. 5J). Moreover, co-IP combined with Western blot revealed that up-regulation of LINC00022 induced the ubiquitination of p21 protein (Fig. 5K). Further, we examined the protein expression and ubiquitination level of p21 in xenografts tumors derived from our in vivo experiment (Fig. 3C), and found that the level of p21 protein decreased and its ubiquitination increased in the OE-022 group compared with the OE-NC group (Supplementary Fig. 8B), consistent with the results of cell experiments. Collectively, these data above suggest that LINC00022 de-stabilizes p21 protein by promoting its decay via UPP.

Both genomic and epigenetic alterations can cause the dysregulation of LncRNA expression in cancer cells [14, 43]. To explore the mechanism responsible for the up-regulation of LINC00022 in ESCC, we first analyzed the copy number variation (CNV) levels of LINC00022 gene and its promoter DNA methylation level in tumor and normal samples from the TCGA database. We showed that the genome CNV level of LINC00022 was positively correlated with its expression in ESCA samples (Supplementary Fig. 9A). In addition, tumor samples with high expression of LINC00022 had higher CNV levels (Supplementary Fig. 9B), and the LINC00022-amplified tumor samples showed higher expression of LINC00022 (Supplementary Fig. 9C). However, there was no significant correlation between DNA methylation in LINC00022 promoter and its transcription level in samples from TCGA-ESCA (Supplementary Fig. 9D-F). These data suggest that the increase of DNA copy number may be one of the genetic mechanisms for the up-regulation of LINC00022 expression.

Recent advances in epigenetic studies have revealed the pivotal role of m6A modification at the post-transcriptional level in RNA metabolism and function in tumor cells [44, 45]. We wondered whether m6A modification was related to the up-regulation of LINC00022 in ESCC. For the first time, we examined the overall m6A levels of RNAs in ESCC using the colorimetric m6A detection assay. Both ESCC cell lines and tumor tissues showed lower levels of RNA m6A compared to the counterparts (Fig. 6A-B), implying that m6A may be involved in the pathogenesis of ESCC. Two m6A online analysis software, SRAMP and RMBase V2.0, predicted 11 and 8 highly reliable m6A modification sites on the LINC00022 transcript (Supplementary Fig. 10A), respectively, among which 3 overlapped loci with high confidence were located at 1074 bp (site1), 1374 bp (site2) and 1462 bp (site3) from the 5′-end (Supplementary Fig. 10B). Subsequently, MeRIP combined with qRT-PCR analysis uncovered that the transcripts of LINC00022 in cell lines Het-1A, KYSE70 and KYSE150 were enriched with m6A at different levels. The m6A modification of site1 was detected in Het-1A and KYSE70 cells, while the m6A modification of site3 was not observed in Het-1A, KYSE70 and KYSE150 cells. Importantly, the m6A modification of site2 was detected in all these three cell lines. The enrichment level of m6A was the highest in Het-1A and the lowest in KYSE150 (Fig. 6C), which is contrary to LINC00022 expression in these three ESCC cell lines, suggesting that m6A modification may play an important role in the regulation of LINC00022.

To further define which m6A enzyme is involved in the regulation of LINC00022, we conducted bioinformatics analysis using the online tool m6A2Target based on sequencing validation data. We found that METTL3, KIAA1429 and FTO may interfere with the expression of LINC00022 (Supplementary Fig. 11A). Further analysis using GEPIA showed that among the three m6A regulators, only FTO was highly expressed in ESCC and was positively correlated with LINC00022 in 182 tumor samples (Supplementary Fig. 11B-D). Therefore, we focused on the role of FTO in ESCC and its regulatory effect on LINC00022. We then analyzed the mRNA expression of FTO in our in-house cohort containing 44 pairs of ESCC specimens by qRT-PCR. FTO was up-regulated in ESCC tumors and positively correlated with LINC00022 expression (Fig. 6D-E). Moreover, the elevation of FTO and its positive correlation with the expression of LINC00022 were also found in a panel of ESCC cell lines (Fig. 6F-G). We used a Lentivirus-mediated system to specifically reduce or increase the intracellular level of FTO in ESCC cells and verified the efficiencies of FTO manipulation by qRT-PCR and Western blot (Supplementary Fig. 12A-B). In KYSE150 and TE1 cells, FTO knockdown resulted in a significant down-regulation of LINC00022 expression (Fig. 6H). In KYSE70 cells, LINC00022 expression was evidently increased after FTO over-expression (Fig. 6I). Neither deletion nor up-regulation of LINC00022 significantly affected the expression level of FTO in ESCC cells (Supplementary Fig. 12C-D), suggesting that FTO is an upstream molecule of LINC00022.

The colorimetric m6A detection assay revealed that knockdown or over-expression of FTO resulted in the increase or decrease of the overall m6A modification level in ESCC cells (Fig. 6J-K), respectively, which further indicated the key role of FTO in the epigenetic regulation of ESCC. Considering that FTO positively regulates the expression of LINC00022, MeRIP combined with qRT-PCR was performed to investigate whether FTO mediated the m6A modification of LINC00022 transcript at site2 in ESCC cells. The results showed that ectopic expression of FTO in KYSE70 cells significantly reduced the enrichment of m6A at site2 of the LINC00022 transcript (Fig. 6L). Both loss-of-function and gain-of-function experiments demonstrated that FTO was an oncogenic factor to facilitate proliferation (Supplementary Fig. 12E-F) and cell-cycle progression (Fig. 6M-N, Supplementary Fig. 12G-H) in ESCC cells, similar to the function of LINC00022. Clinically, survival analysis from TCGA-ESCC cohort depicted those patients with high FTO expression had a shorter median survival than those with low FTO expression (Supplementary Fig. 13A, HR = 1.43, p = 0.38, 25.4 months vs. 42.1 months). We also analyzed the relationship between FTO and clinical parameters of 36 patients with complete information in our in-house cohort. FTO was up-regulated in tumor tissues of female patients or advanced stage patients, compared with male or early-stage patients (Supplementary Fig. 13B), respectively. No significant correlation was found between FTO expression and age, differentiation, grade, lymph node metastasis or depth of invasion (Supplementary Fig. 13C). Taken together, these data suggest that up-regulated FTO mediates the RNA m6A demethylation of LINC00022 in ESCC cells and promotes its expression, thus, enhancing ESCC cell proliferation.

The m6A readers play crucial roles in determining the fate of m6A-marked RNAs, such as stabilization and degradation [46]. It has been found that, unlike other m6A readers that stabilize RNA, YTHDF2 tends to de-stabilize m6A-modified RNA molecules and facilitates their degradation [47, 48]. Our above results showed that the expression level of LINC00022 was increased in ESCC cells (Fig. 1F), while its m6A modification level was decreased (Fig. 6C). We therefore hypothesized that YTHDF2 might regulate the m6A-modified LINC00022 transcript in ESCC cells. To confirm this hypothesis, we first determined the protein levels of YTHDF2 in 14 pairs of ESCC tumor specimens and the matched normal specimens by Western blot, and found that YTHDF2 was decreased in 13 of the 14 cases of ESCC tumors (Fig. 7A). Pearson’s coefficient analysis showed the inverse correlation between YTHDF2 and LINC00022 expression in 14 cases of ESCC tumors (Fig. 7B). Lower YTHDF2 levels were also detected in KYSE70 and KYSE150 cell lines, compared with Het-1A (Fig. 7C). RIP analysis confirmed the direct binding between YTHDF2 protein and LINC00022 transcript in KYSE70 and KYSE150 cells (Fig. 7D). We used siRNAs to down-regulate the intracellular level of YTHDF2 and a plasmid-mediated system to increase the expression of YTHDF2 in ESCC cells (Fig. 7E). The level of LINC00022 in KYSE70 cells was significantly up-regulated, albeit modestly, after YTHDF2 deletion (Fig. 7F). Ectopic expression of YTHDF2 led to an obvious decrease in LINC00022 level in KYSE150 cells (Fig. 7F). Importantly, ablation of YTHDF2 slowed down the decay rate of LINC00022 in ESCC cells, while over-expression of YTHDF2 accelerated the degradation of LINC00022 transcript in the presence of Act D (Fig. 7G-H), an RNA synthesis inhibitor. We next investigated the biological function of YTHDF2 in ESCC, and found that over-expression of YTHDF2 inhibited the proliferation and cell cycle progression of ESCC cells (Supplementary Fig. 14A-B), while knockdown of YTHDF2 had the opposite effect (Supplementary Fig. 14C-D). Clinically, Kaplan-Meier analysis of TCGA-ESCC dataset showed that high level of YTHDF2 was a favorable factor for patient OS (Supplementary Fig. 14E). This indicates that, contrary to LINC00022 and FTO, YTHDF2 plays a tumor suppressor role in ESCC. Furthermore, we confirmed that YTHDF2 deletion rescued the decrease in LINC00022 levels caused by FTO knockdown in KYSE150 cells (Fig. 7I). Enforced YTHDF2 abolished the up-regulation of LINC00022 induced by ectopic FTO expression in KYSE70 cells (Fig. 7J). These findings indicate that FTO promotes LINC00022 up-regulation in ESCC cells via an m6A-YTHDF2-dependent mechanism.

To elucidate the role of the FTO/LINC00022 axis in tumorigenesis of ESCC, we carried out functional rescue experiments in vitro and in vivo. CCK-8 detection showed that knockdown of LINC00022 almost fully abolished the pro-proliferative effect of FTO on ESCC cells (Fig. 8A). Genetic addition of FTO effectively relieved the inhibitory impact of LINC00022 knockdown on the proliferation of ESCC cells (Fig. 8B). The regulation of FTO/LINC00022 cross-talk on ESCC cell growth was further confirmed by colony formation assays (Fig. 8C). PI-labeling flow cytometry revealed that LINC00022 deletion significantly retarded FTO-induced cell-cycle progression in KYSE70 cells (Fig. 8D, Supplementary Fig. 15A) while over-expression of FTO partially alleviated the G0/G1 phase arrest elicited by ablation of LINC00022 in KYSE150 cells (Fig. 8E, Supplementary Fig. 15B). Concomitantly, over-expression of FTO reduced the protein levels of p16, p21, and p53 in KYSE70 cells, and knockdown of LINC00022 counteracted, at least in part, these effects (Fig. 8F). Moreover, ectopic expression of FTO blocked the elevation of p16, p21, and p53 caused by LINC00022 deficiency in KYSE150 cells (Fig. 8G). The xenografts experiment in nude mice further demonstrated that ablation of LINC00022 completely mitigated FTO-induced cell proliferation of KYSE70 in vivo (Fig. 8H, Supplementary Fig. 15C), whereas over-expression of FTO partially relieved the inhibitory effect of tumor formation caused by LINC00022 knockdown (Fig. 8I, Supplementary Fig. 15D). Collectively, these data indicate that the FTO/LINC00022 axis regulates ESCC growth in vitro and in vivo.

The procedures were granted from the Ethical Review Committee of Zhengzhou University.

All authors approved this manuscript for publication in Journal of Experimental & Clinical Cancer Research.

No potential conflicts of interest.

All data analyzed in this study are available from the corresponding author upon reasonable request.