The GC cell lines AGS, MKN28 and MKN45 were obtained from the Type Culture Collection of the Chinese Academy of Sciences (China). The normal gastric epithelial cell line GES-1 was purchased from the Lab Animal Center of the Fourth Military Medical University (China). All cell lines were routinely tested and had negative results for mycoplasma. All cells except AGS were maintained in DMEM (Invitrogen) and AGS in F12 (Invitrogen) supplemented with 10% fetal bovine serum (GIBCO) in an incubator with 5% CO2 and 95% room air at 37°C. The LINC00893 gene was cloned into the pEZ-Lv201 lentivirus vector (GeneCopoeia), RBFOX2 gene was cloned into the pcDNA3.1 vector (GeneCopoeia), and wild-type TP53 were cloned into the pEnter lentivirus vector (Vigene Biosciences). All the small interfering RNAs (siRNA) and oligonucleotides (ASO) were purchased from RiboBio (China). All plasmids, siRNA and ASO were transfected with Lipofectamine 3000 reagent (Invitrogen) following the instructions. The sequences of siRNAs and ASO used are listed in Supplementary Table S5.

RNA sequencing was performed using an Illumina HiSeqTM 2500. Sample preparation was performed based on the manufacturer’s standard protocols. The cDNA/DNA/small RNA libraries were sequenced on the Illumina sequencing platform by Genedenovo Biotechnology Co., Ltd. (Guangzhou, China). The public datasets analyzed in the current study are available in TCGA (https://cancergenome.nih.gov/), GEPIA (http://gepia.cancer-pku.cn/), and KM Plotter (http://kmplot.com/analysis/). GSEA 3.0 (http://www.gsea-msigdb.org/gsea/index.jsp) was used to perform Gene set enrichment analysis (GSEA).

Strand-specific genomic coordinates for all exons of human LINC00893, PINCR, NEAT1, GAPDH, SDHA and UBC genes were downloaded from the UCSC Genome Browser (GRCh38/hg38) in BED format. PhyloCSF was applied to generate FASTA alignments to assess the coding potential (the codon substitution frequencies score - CSF) of individual exons and of mature transcripts of the above genes. The Coding Potential Assessment Tool (CAPT, http://lilab. research.bcm.edu/cpat/) constructs a logistic regression model built with four sequence features: open reading frame coverage, open reading frame size, hexamer usage bias and the Fickett TESTCODE statistic. The cutoff value for human coding probability (CP) was set at 0.364. CP < 0.364 was considered a noncoding transcript, whereas CP ≥ 0.364 was defined as a coding transcript.

RNA isolation plus (TaKaRa, Japan) was used to extract total RNA from tumor tissues or cells. Single-stranded cDNA was generated from 1 μg of total RNA in a 20 μL reaction volume system using oligodT primers according to the protocol supplied with Primer ScriptTM RT Reagent (TaKaRa, Japan). The relative expression levels were measured by qPCR using an ABI 7900HT instrument (Applied Biosystems, United States) in a total volume of 10 µL with the SYBR green detection system (Takara, Japan), and GAPDH was used as an endogenous control. The primers used for the qPCR analysis are listed in Supplementary Table S5.

Cell viability was examined by Cell Counting Kit-8 (CCK-8; Dojindo Laboratories, Japan). Briefly, after treatments, CCK-8 solution (10 μL) was added to each well, and the plates were incubated for an additional 2 h. The absorbance of each well was measured at 450 nm using a microplate reader (MultiskanEX, Lab systems, Finland). In the colony formation assay, treated cells (1×10³ cells/well) were seeded into 6-well plates and cultured in DMEM/F12. Then the medium was replaced every 3 days. After 14 days, colonies were fixed with paraformaldehyde and stained with 0.1% crystal. After that, colonies were counted.

Firstly, GC cells were transfected as described above for 48 h. Next, cells were harvested and counted. Then a certain number of transfected cells (5×10⁴ cells/well) were cultured with serum-free medium in the upper chamber of Transwell plates (Corning, United States). Medium containing fifteen percent FBS was added into the lower chamber of the Transwell plates. For the invasion experiments, the upper chamber was covered with DMEM (F12) and Matrigel (BD Biosciences, CA) mixture in advance. Finally, cells on the top of the chamber were wiped with cotton swabs, while cells that went through the membrane were stained with 0.1% crystal violet, observed and counted under microscope at ×100 magnification.

Four-week-old female athymic BALB/c nude mice were fostered under specific pathogen-free conditions and treated according to protocols. Lv201-LINC00893 or control vector were stably transfected into MKN45 cells. To assess the cell proliferation viability in different groups in vivo, the stable overexpressing cells were harvested, washed with cold PBS and resuspended to 5 × 10⁷ cells/ml. Then, 100 μL cell suspension was subcutaneously injected into the armpits of each mouse. Tumor growth was examined and recorded every week, and tumor volumes were calculated using equation V = 0.5 × D × d² (V, volume; D, longitudinal diameter; d, latitudinal diameter). Four weeks after injection, mice were sacrificed, and two groups of subcutaneous tumors were obtained and imaged separately. In the cell metastasis assay in vivo, the transfected cells were resuspended at a concentration of 1 × 10⁷ cells/ml, and then 100 μL of resuspended cells was injected into the tail veins of the two groups of mice. Eight weeks after injection, the mice were euthanized. Lungs of the mice in two groups were removed and photographed. We then performed hematoxylin and eosin (HE) staining after tissue fixation and sectioning to evaluate the metastatic nodes, besides, the number of metastatic nodes in each lung was counted. The animal experiments were conducted in accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. Our whole protocol was approved by the Committee on the Ethics of Animal Experiments of the First Affiliated Hospital of Sun Yat-sen University.

Cytoplasmic and nuclear fractions of RNA were isolated in accordance with the instructions of PARIS™ Kit (AM1556, Thermo Fisher Scientific, United States). Briefly, AGS and MKN28 cells were washed with cold PBS and lysed in Cell Fraction Buffer on ice for 10 min. Then gathering the fractions into new tubes and centrifugating at 500 × g for 3 min at 4°C. After that, the supernatant was collected as the cytoplasmic fraction. Finally, washing the rest pellet with Cell Fraction Buffer and collecting the nuclei.

RNA was extracted from AGS cells or transcribed in vitro and purified. Purified RNA was then incubated with specific prewashed beads at 37°C for 30 min. After centrifugation, the supernatant was hybridized with denatured biotin-labeled RNA probes at 55°C for 2 h. Next, we added prewashed beads, incubated them at 37°C for 1 h, centrifuged them, and washed the beads. The pulldown RNA was extracted and analyzed by quantitative reverse transcription PCR (RT-qPCR). LINC00893 RNAs were purified, and equal amounts of transcripts were used in the binding assays. The binding components were pulled down by biotin-labeled RNA probes and analyzed by western blotting.

RNA immunoprecipitation (RIP) assays were performed using a Magna RIP RNA-Binding Protein Immunoprecipitation Kit (Millipore, United States) according to its instructions. Anti-RBFOX2 (Bethyl, United States) and normal Rabbit IgG (CST, United States) were used to immunoprecipitated with target RNA or as a negative control. Finally, RT-qPCR analysis was performed to detect the enrichment of the coprecipitated RNA binding to RBFOX2.

Cy3 labeled LINC00893 probe was designed and synthesized by GenePharma (Shanghai, China). For FISH assay, AGS or MKN28 cells (1 × 10⁵) were fixed in 4% formaldehyde and permeabilized with 0.3% Triton X-100 for 15 min, washed with PBS three times and once in 2× SSC buffer. Hybridization was carried out using DNA probe sets at 37 °C for 16 h. Images were obtained with confocal laser scanning microscope (Zeiss, Germany) and processed using the ZEN imaging software.

Total proteins were extracted using RIPA supplemented with protease and phosphatase inhibitor reagents (Thermo-Fisher Scientific, United States). Equal amounts of protein samples were separated by SDS/PAGE and then electrophoretically transferred to PVDF membranes (Millipore, United States). Next, the membranes were probed with specific primary antibodies against p53 (dilution, 1:1,000; Abcam, United States), p21 (dilution, 1:1,000; CST, United States), RBFOX2 (dilution, 1:1,000; Bethyl, United States), E-cadherin (dilution, 1:1,000; CST, United States), N-cadherin (dilution, 1:1,000; CST, United States), vimentin (dilution, 1:1,000; CST, United States) and GAPDH (dilution, 1:5,000; Proteintech, China) at 4°C overnight, followed by incubation with HRP-conjugated secondary antibodies (dilution, 1:5,000; CST, United States) at room temperature for 1 h. Immunoreactive bands were visualized using enhanced chemiluminescence reagents (Bio-Rad, United States), and GAPDH was considered an inner loading control.

All data are presented as the mean ± standard deviation, appropriate statistical methods including Student’s t-test, Wilcoxon signed-rank test, Mann-Whitney test, Pearson chi-square test were used to calculate differences between groups, and p < 0.05 indicates statistical significance. Statistical analyses were performed using SPSS Statistics software (version 18.0) or GraphPad Prism 7 software.

First, to screen out the potential tumor suppressor lncRNAs in GC, we performed RNA sequencing of three pairs of human GC tissues and matched nontumor tissues. Differentially expressed lncRNA transcripts were identified. A total of 409 lncRNA transcripts were found to have significant decreases in GC tissues vs nontumor tissues (Fold change<0.5, p < 0.05), while 385 were found to have significant increases. (Fold change>2, p < 0.05) (Figure 1A; Supplementary Table S1). Nutlin-3 is an inhibitor of human homolog of murine double minute 2 (MDM2), which is a negative regulator of p53 (Endo et al., 2011). AGS is a TP53-wild type GC cell line, in which nutlin-3 has been proved to significantly elevate the expression of p53 (Wei et al., 2010; Vaseva et al., 2011). To determine the optimal concentration and time of nutlin-3 stimulation, we treated AGS cells with gradient concentrations and time periods of nutlin-3 and then detected the expression of p53 and its downstream effector p21 (Supplementary Figure S1A–D). Afterwards, we treated AGS cells with nutlin-3 at the most appropriate concentration (10 µM) and time (12 h) and then conducted RNA sequencing to identify the differentially expressed lncRNA transcripts (Figures 1B,C). As the cluster map shows, 152 transcripts were downregulated when AGS was stimulated with nutlin-3 (Fold change<0.5, p < 0.05), while 146 transcripts were upregulated, which were considered potential targets of p53 (Fold change>2, p < 0.05) (Figure 1D; Supplementary Table S2). Next, we cross-analyzed the lncRNAs differentially expressed in gastric tissues and AGS cells elevated after the accumulation of p53, and we found there were 40 lncRNAs differentially expressed in both sequencing data (Figure 1E). Intriguingly, there were only four lncRNAs decreased in GC tissues (Fold change<0.5, p < 0.05) and, at the same time, upregulated in AGS cells treated with nutlin-3 (Fold change>2, p < 0.05) (Figure 1F). This might be due to the highly heterogeneous nature of GC tissues and huge differences of the lncRNA expression profiles between GC tissues and AGS cells. We used GEPIA (Tang et al., 2017) to analyze the expression of these four candidate lncRNAs (Figure 2A, Supplementary Figure S1E). In addition, RT-qPCR confirmed that the level of LINC00893 was significantly decreased in fresh gastric tumor tissues compared with nontumor tissues, while there were no differences in the other three lncRNAs (Figure 2B, Supplementary Figure S1F). We also found there were no significant differences between the expression of these three lncRNAs and the survival of GC patients (Supplementary Figure S1G). Therefore, we conducted further research on LINC00893. The noncoding capability of LINC00893 was confirmed by analysis with PhyloCSF (Lin et al., 2011) and the Coding Potential Assessing Tool (CPAT) (Duan et al., 2021) (Figure 2C, Supplementary Figure S1H).

In addition, the correlation of the expression of LINC00893 and the clinicopathological characteristics of patients was analyzed. As shown in Table 1, LINC00893 expression was negatively associated with tumor size, primary tumor (T) stage, lymph node metastasis, distant metastasis and tumor-node-metastasis (TNM) stage. Further analysis indicated that low expression of LINC00893 was negatively correlated with worse survival of GC patients (Figure 2D), suggesting that LINC00893 may inhibit GC progression.

To delineate the possible link to p53, we looked for a correlation between LINC00893 and p53. First, we found that nutlin-3 treatment significantly increased the expression of LINC00893 in AGS cells, which express wild-type (WT) p53 (Figures 2E,I). Next, we treated AGS cells with the chemotherapy drug 5-fluorouracil (5-FU), which could cause DNA damage and increase the level of p53. We found that LINC00893 increased with higher concentration gradients, consistent with p53 (Figures 2F,J). To rule out other effects of nutlin-3 and 5-FU, we transfected TP53 plasmids or siRNAs and verified the expression of p53 (Figures 2G,H). With the overexpression of p53, LINC00893 was significantly elevated, while it was reduced after the knockdown of p53 (Figures 2K,L). Despite its p53-dependent induction, we were unable to identify an obvious p53 binding site in the vicinity of the LINC00893 promoter, nor was one identified in a recent p53 ChIP-seq study performed in TP53 WT cells (Sanchez et al., 2014), thus suggesting that p53 may regulate LINC00893 expression via an indirect pathway, possibly through other p53 targets associated with transcription or p53 binding to enhancer regions. To identify promoter elements involved in p53-mediated regulation of LINC00893, we then analyzed the promoter of LINC00893 and predicted the potential transcription factors (TFs). Among the predicted TFs, Sp1 transcription factor (SP1) was predicted both in GeneCards and PROMO (Farre et al., 2003). Intriguingly, SP1 has been proved to be regulated by p53 and act as an intermediary TF between p53 and its target gene (Innocente and Lee, 2005; Li et al., 2014a). Therefore, we speculated that p53 regulates LINC00893 expression through p53/SP1/LINC00893 axis.

According to the correlation between LINC00893 expression and GC, we focused on the biological functions of LINC00893 in GC proliferation and metastasis. AGS is an epithelial cell line derived from gastric adenocarcinoma and MKN28 is a highly metastatic GC cell line, which are widely used as models to detect the roles of oncogene or tumor suppressor gene in GC in vitro. The expression vector Lv201-LINC00893 or ASO-LINC00893 was transfected into GC cells, and the efficiency of LINC00893 overexpression or knockdown was confirmed by RT-qPCR (Figure 3A). CCK8 and colony formation assays were conducted, and they showed strikingly inhibited proliferation ability when LINC00893 was overexpressed or a promoting proliferation ability when LINC00893 was knocked down (Figures 3B,C). Transwell assays showed that LINC00893 overexpression significantly decreased the migration and invasion potential of GC cells, while LINC00893 knockdown showed the opposite effects (Figures 3D,E). To explore the downstream signaling pathways involved in LINC00893, we performed RNA-seq and bioinformatic analysis in AGS cells treated with empty vector or LINC00893 overexpression vector. Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment and Gene Set Enrichment Analysis (GSEA) results indicated that LINC00893 was correlated with pathways directly involved in the regulation of cell adherens junctions (Supplementary Table S3; Figure 3F; Supplementary Figure S1I). Therefore, we detected the key representative proteins of EMT. We found that overexpression of LINC00893 increased the level of E-cadherin and decreased the levels of N-cadherin and vimentin. The opposite effect was confirmed after knockdown of LINC00893 (Figure 3G). This suggests that LINC00893 can inhibit the process of EMT in GC cells.

To determine the roles of LINC00893 in GC tumorigenesis in vivo, MKN45, a GC cell line derived from gastric lymph nodes in a female patient with signet ring cell carcinoma, was used to construct model cells stably overexpressing LINC00893 or the control vector. Then, cells were subcutaneously injected into nude mice (Figure 4A). We found mice in the LINC00893-overexpressing group generated smaller tumors than the control group (Figure 4B), and HE staining was performed to identify the tumors (Figure 4C left panel). Tumor volume and weight also indicated that the overexpression of LINC00893 suppressed tumor growth in vivo (Figures 4D,E). Besides, a significant difference in Ki-67 staining was found between the LINC00893 and control group (Figure 4C right panel). Furthermore, by injecting the stably transfected MKN45 cells into tail veins of nude mice, it was found that LINC00893 dramatically inhibited the development of pulmonary metastasis (Figures 4F–H).

Several recent studies have revealed the closely related regulatory network between lncRNAs and proteins. GO (Gene Ontology) enrichment indicated that LINC00893 might play a role though binding to proteins (Supplementary Figure S2B). First, to identify the subcellular localization of LINC00893, we detected LINC00893 expression in cytoplasmic and nuclear fractions by RT-qPCR analysis (Figure 5A). The results showed that LINC00893 was localized predominantly in the nucleus, with some localization in the cytoplasm, which was further verified by RNA fluorescence in situ hybridization assays (Figure 5B). The secondary structure of LINC00893 was also predicted (Figure 5C). By using the StarBase database (Li et al., 2014), a large number of proteins were predicted to directly bind to LINC00893 (Supplementary Table S4). The splicing factor RNA Binding Protein Fox-1, Homolog 2 (RBFOX2) was included, which is a classic nuclear regulator of Polycomb Repressive Complex 2 (PRC2) recruitment to active genes, particularly those with bivalent histone modifications, to mediate dynamic transcriptional control and regulate autoregulation in RNA-binding protein networks (Venables et al., 2013; Braeutigam et al., 2014; Jangi et al., 2014; Wei et al., 2016). We found that nucleotides 753-848 of the LINC00893 sequence and amino acids 126-177 of the RBFOX2 sequence were the most likely binding areas (Figure 5D). To verify the direct interaction between LINC00893 and RBFOX2, western blotting was carried out using antibodies specific for RBFOX2 to test the protein samples precipitated by the LINC00893 CHIRP assay. This result indicated that RBFOX2 was pulled down by LINC00893 (Figures 5E,F). To further confirm this interaction, we carried out an RIP assay using a RBFOX2-specific antibody, followed by RT-qPCR using primers specific for LINC00893. As expected, LINC00893 was enriched in the anti-RBFOX2 group compared to the control IgG group (Figure 5G). These results indicated that RBFOX2 directly interacts with LINC00893.

Then, we examined RBFOX2 expression when LINC00893 levels were altered to determine whether LINC00893 regulated RBFOX2 expression. Interestingly, LINC00893 overexpression reduced the levels of the RBFOX2 protein (Figures 5H,I). According to our findings above, we inferred that LINC00893 regulated RBFOX2 protein stability. To elucidate the underlying mechanism by which LINC00893 regulated the RBFOX2 protein, we performed cycloheximide chase assays and observed an obvious shorter half-life of RBFOX2 in LINC00893-overexpressing cells than in empty control cells (Figure 5J). The ubiquitin-proteasome degradation pathway has been reported to be a primary mechanism regulating protein levels in cells. The proteasome inhibitor MG132 antagonized the decrease in RBFOX2 protein levels caused by LINC00893, indicating that LINC00893 downregulated the expression of RBFOX2 by promoting its proteasomal degradation (Figure 5K). Subsequent ubiquitination assays revealed that LINC00893 overexpression increased the level of the ubiquitinated RBFOX2 protein in AGS and MKN28 cells (Figure 5L). Taken together, LINC00893 directly interacts with RBFOX2 and promotes its ubiquitin-mediated degradation in GC cells.

RBFOX2 promotes tumor progression by acting as a splicer of EMT-related genes in several types of cancers (Shapiro et al., 2011; Braeutigam et al., 2014); however, the status and role of RBFOX2 in GC is unclear. Therefore, we focused on how RBFOX2 affects EMT-related genes and its role in GC cells. First, we knocked down RBFOX2 with siRNA or increased it by transfecting an overexpression plasmid. We found that the proliferation ability of GC cells was greatly decreased when RBFOX2 was downregulated but increased when RBFOX2 was upregulated (Figures 6A,B). The migration and invasion abilities of GC cells showed the same tendency (Figures 6C,D). Additionally, we found that when RBFOX2 was knocked down, the expression of E-cadherin increased, while N-cadherin and vimentin decreased, in accordance with the results when RBFOX2 was overexpressed (Figure 6E). In addition, analysis from KM Plotter (http://kmplot.com/analysis/) showed that high expression of RBFOX2 was correlated with a worse survival of GC patients, which is the opposite tendency with LINC00893 in terms of the overall survival rate (Figure 6F).

Subsequently, we determined whether LINC00893 exerts its function though RBFOX2. We found that the LINC00893-mediated suppression of proliferation, migration and invasion was remarkably abolished when we overexpressed LINC00893 and RBFOX2 in GC cells at the same time. Consistent with these results, transfecting siRNA of RBFOX2 and ASO of LINC00893 simultaneously strikingly eliminated the promotion of GC cell proliferation, migration and invasion induced by LINC00893 knockdown, suggesting that LINC00893-mediated RBFOX2 downregulation contributes to the suppression of LINC00893 on GC cell proliferation, migration and invasion (Figures 7A–D). In addition, the changes of the EMT-related proteins N-cadherin, vimentin and E-cadherin caused by the upregulation or downregulation of LINC00893 was rescued by the overexpression or knockdown of RBFOX2, respectively (Figure7E).