N6-methyladenosine-dependent pri-miR-17-92 maturation suppresses PTEN/TMEM127 and promotes sensitivity to everolimus in gastric cancer

N6-methyladenosine (m6A) is the most common epigenetic RNA modification with essential roles in cancer progression. However, roles of m6A and its regulator METTL3 on non-coding RNA in gastric cancer are unknown. In this study, we found elevated levels of m6A and METTL3 in gastric cancer. Increased METTL3 expression indicated poor outcomes of patients and high malignancy in vitro and in vivo. Mechanically, m6A facilitated processing of pri-miR-17-92 into the miR-17-92 cluster through an m6A/DGCR8-dependent mechanism. The m6A modification that mediated this process occurred on the A879 locus of pri-miR-17-92. The miR-17-92 cluster activated the AKT/mTOR pathway by targeting PTEN or TMEM127. Compared with those with low levels of METTL3, METTL3-high tumors showed preferred sensitivity to an mTOR inhibitor, everolimus. These results reveal a perspective on epigenetic regulations of non-coding RNA in gastric cancer progression and provide a theoretical rationale for use of everolimus in the treatment of m6A/METTL3-high gastric cancer.


Introduction
Gastric cancer is one of the most common malignancies and the third leading cause of cancer-related death worldwide 1 . With few specific symptoms in early stages and with low rates of gastroscopy, most patients have already reached an advanced stage at the time of initial diagnosis. Even among patients who underwent curative resection, 60% suffered recurrences and distant metastasis, with a median overall survival (mOS) of <12 months 2 . Moreover, gastric cancers with peritoneal metastasis respond rarely to any treatments, leading to an extremely inferior prognosis with life expectancy <6 months 3 . The effect of targeted therapy is quite limited by the lack of dominant driver genes in gastric cancer. Trastuzumab is the only target drug approved for the first-line treatment of advanced gastric cancer based on the TOGA trial, but its usage was confined in a small part of the patients with ERBB2 amplification 4 . The antiangiogenic drug bevacizumab only improves overall survival in non-Asian patients as the first-line treatment 5 . Up to now, the immune checkpoint inhibitors, pembrolizumab (KEYNOTE-059 cohort 1) 6 and nivolumab (ATTRACTION-02) 7 , are only approved for third-line and later treatment in gastric cancer, with response rates <15%. Therefore, an in-depth investigation of the molecular mechanisms in gastric cancer oncogenesis and progression is critical to allow early diagnosis, innovative therapeutic methods, and ultimately improved prognosis and quality of life for patients. N 6 -Methyladenosine (m 6 A) is the most common epigenetic modification in eukaryotic messenger RNA (mRNA) 8 and non-coding RNA (ncRNA) 9 . It plays a crucial role in gene expression by participating in almost every stage of mRNA metabolism and exerts vital and specific roles in the pathogenesis of various cancers 10 . Recently, several studies reported that METTL3, the core methyltransferase for m 6 A modification, promotes gastric cancer progression 11,12 . However, these studies only focused on m 6 A of mRNA, and few studies investigated m 6 A of ncRNA in gastric cancer or implied the possible clinical translational value of m 6 A/METTL3 related signaling pathways.
Here, we investigated the biological function of METTL3/m 6 A in regulating ncRNA and defined a novel pathway for m 6 A-dependent primary microRNA (miRNA) maturation and AKT/mTOR activation in gastric cancer, which could be counteracted by everolimus.

Clinical samples
Fresh tissues were obtained from gastric cancer patients who underwent radical resections at Qilu Hospital of Shandong University between January 2017 and August 2018. Pathologically confirmed gastric cancer paraffinembedded tissues between 2009 and 2014 were obtained from the Department of Pathology, Qilu Hospital of Shandong University. Only patients with evidence of survival and recurrence were included for OS and RFS analysis, respectively.

RNA m 6 A quantification
Total RNA was isolated from 10 mm 3 of fresh tissue or 10 6 cells using TRIzol (Invitrogen, CA, USA). The m 6 A RNA Methylation Quantification Kit (Abcam, Cambridge, UK) was used to quantify the m 6 A content according to the manufacturer's instructions. The optical absorbance was measured by a SpectraMax Plus384 Microplate Spectrophotometer (Molecular Device, Sunnyvale, CA, USA).

Immunohistochemistry (IHC)
In clinical studies, paraffin-embedded sections were blocked by goat serum and stained with anti-METTL3 antibody (Abcam, Cambridge, UK) using an IHC staining kit (Zsbio, Beijing, China). Cell nuclei were stained with hematoxylin. In animal studies, tumor xenografts or peritoneal tumors were fixed and processed with a similar procedure with anti-METTL3, anti-Ki67 (Abcam, Cambridge, UK), and anti-PTEN antibody (CST, Danvers, MA).

Proliferation and colony-formation assays
For proliferation assays, cells were seeded in 6-well plates (10,000 or 20,000 cells per well) and counted every 24 h. For colony-formation assays, suspended single cells were seeded in 6-well plates (1000 cells per well), and colonies were counted within 14 days.

Wound-healing, migration, and invasion assays
For wound-healing assays, wounds were made by scratching a line using a 200 µL tip, and the intervals were measured within 72 h. For migration and invasion assays, a Transwell system (Corning, NY, USA) was used as previously described 13 . Migrated and invaded cells were stained with crystal violet (Beyotime, Shanghai, China) and photographed.

Subcutaneous xenograft and peritoneal implant models
Six-week female BALB/c Nude Mice were purchased from Vital River Laboratory (Beijing, China). For subcutaneous xenograft models, 0.1 mL of cell suspension containing 10 6 cells were injected subcutaneously into the right flank of mice (n = 6 for each group). Mice were sacrificed at 21-or 32-day after injection. For peritoneal implant models, a cell suspension (5 × 10 6 cells) was injected intraperitoneally. All mice were sacrificed 4 weeks after injection (n = 4 for each group). Mice were randomly allocated into each group with no blinding. The animal studies were performed following the ARRIVE guidelines and were approved by the Animal Ethical Committee of Qilu Hospital of Shandong University.

RNA immunoprecipitation (RIP)
A Magna RIP RNA-Binding Protein Immunoprecipitation Kit (Millipore, Darmstadt, Germany) was used for RIP. Briefly, cells were lysed and mixed with anti-m 6 A (Abcam, Cambridge, UK), anti-DGCR8 (Abcam, Cambridge, UK) antibodies, or isotype controls (Abcam, Cambridge, UK). The antibody-binding RNA was pulled down by protein A/G magnetic beads and quantified by real-time PCR.

Everolimus-sensitivity assays
For the in vitro study, everolimus (APExBIO, Houston, TX, USA) was added to cells with final concentrations of 5 or 50 µg/mL. Cell viability was measured using a CCK-8 kit (BestBio, Shanghai, China). For the in vivo study, subcutaneous xenograft models with METTL3-high (n = 3) and control cells (n = 3) were established as above. Control cells were inoculated two days before METTL3-high cells. When the tumor sizes were similar, volume-matched mice received everolimus (50 µg/day intragastrically) or solvent for 17 days and were sacrificed at day 18. Mice were randomly allocated to each group with no blinding.

In silico analyses
All datasets used in this study were derived from public databases. Enrichment analyses were performed by DAVID (https://david.ncifcrf.gov) and TAM (http://www. cuilab.cn/tam). Prediction of miRNA targets was performed by the online tool miRDB (http://mirdb.org/).

Statistical analyses
Data were from at least three independent experiments unless otherwise specified. Variations within each group were estimated and were all statistically compared. Differences between groups were calculated by Student's ttests. Univariate analyses were performed by Chi-squared tests. Survival data were compared by log-rank tests. Correlations were analyzed by linear regression. All the statistical tests were two-tailed. P values < 0.05 were considered statistically significant. Data were presented as mean or mean ± standard deviation (SD).

Results
Elevated m 6 A is mainly regulated by its "writer" METTL3 in gastric cancer To explore the features of m 6 A in gastric cancer, we first compared the levels of m 6 A on total RNAs from 12 pairs of cancerous and adjacent tissues. The m 6 A levels were significantly elevated in tumor tissues compared with their adjacent tissues (Fig. 1a). To determine the key regulators, we performed differential expression analysis on the m 6 A "writers", "erasers", and "readers" from gastric cancer and normal tissues in The Cancer Genome Atlas Stomach Adenocarcinoma (TCGA-STAD) database. The "writers" and "readers" were all overexpressed in gastric cancer, whereas "erasers" were expressed similarly in tumors and controls (Fig. 1b). Based on this, we then focused on the most differentially expressed "writer", METTL3, for further study.
METTL3 RNA level was significantly higher in cancerous tissues than that in adjacent tissues (Fig. 1c). In addition, METTL3 RNA positively correlated with total RNA m 6 A levels in our cohort (Fig. 1d). When mutations were considered in the previously published cohorts, genetic changes in METTL3 only accounted for 2.12% of total patients (Fig. 1e).

METTL3 overexpression indicates adverse pathological features and poor outcome in gastric cancer
To determine the clinical relevance of METTL3, we tested METTL3 expression in gastric cancer samples of 87 patients who had received radical or palliative gastrectomy (Fig. 1f). METTL3 was mainly distributed in the nucleus of tumor cells. According to the staining intensity, 12 negative (−), 24 weak positive (+), 32 medium positive (++), and 19 strong positive cases (+++) were observed. METTL3 expression significantly correlated with AJCC staging (the eighth edition, P = 0.0056), lymph node metastasis (P = 0.0251), and vascular invasion (P = 0.0346), but did not correlate with gender, age, differentiation, and tumor invasion (Table 1   d Correlation between METTL3 mRNA levels (y-axis) and percentage of m 6 A content in total RNA (x-axis) in our clinical cohorts (n = 24). e Genetic changes of METTL3 in gastric cohorts from online databases. Original data were extracted from cBioPortal. f Representative IHC staining of METTL3 in GC tissues (100×). One representative sample of staining intensity (-), (+), (++), and (+++) were shown. Scale bars represent 50 μm. g Clinicopathologic features of patients with METTL3-low (n = 36) and high tumors (n = 51). h, i Kaplan-Meier plots of RFS (n = 58) and OS (n = 80) of patients with METTL3-low and high tumors according to IHC staining intensity. Data were derived from a gastric cancer cohort followed-up by us. j, k Kaplan-Meier plots of RFS (n = 641) and OS (n = 876) of patients with METTL3-low and high tumors according to mRNA levels. Data were interrogated from the GEO datasets GSE14210, GSE15459, GSE22377, GSE29272, GSE38749, GSE51105, and GSE62254 with the probe 209265_s_at. Data are presented as mean ± SD. *P < 0.05; **P < 0.01.
Next, we established a gastric cancer peritoneal metastasis model by intraperitoneal injection of tumor cells (Fig. 3g). The abdominal circumferences of the mice bearing METTL3-high cells increased rapidly and became larger than those of the mice bearing control cells from the second week (all P < 0.05, Fig. 3h). However, body weights were not significantly different between the two groups ( Fig. 3i). Most mice with METTL3-high cells showed palpable masses on the abdominal wall, which was not observed in the control group (Fig. 3g). When dissected four weeks later, the mice bearing METTL3high cells possessed more and larger peritoneal-implanted nodules (both P < 0.0001, Fig. 3j, k). They were mainly distributed in the mesentery and omentum (Fig. 3j), while few grew on the surface of the liver or spleen. METTL3 overexpression in implanted peritoneal nodes was confirmed by IHC (Fig. 3l).
We then examined the expression levels of the miR-17-92 cluster and pri-miR-17-92 in cells and xenograft tumors. Both in vitro or in vivo, METTL3 overexpression significantly increased levels of all six miRNAs (all P < 0.05, Fig. 4e) and reduced the level of pri-miR-17-92 (P = 0.0229 in vitro and P = 0.0009 in vivo, Fig. 4f), while METTL3 downregulation reduced all miRNAs (all P < 0.05, Fig. 4g) and increased the level of pri-miR-17-92 (P = 0.0078 in vitro and P = 0.0009 in vivo, Fig. 4h).
The correlation between METTL3 and the miRNA-17-92 cluster was also confirmed in our clinical samples. Levels of miR-17, 18, 19a, 19b-1, and 20a, as well as the average level of all six miRNAs (miR-mean), were significantly elevated in tumors compared with those in adjacent tissues (P < 0.05 for all, Fig. 4i). The expression of each member of the miRNA-17-92 cluster was positively correlated with the others (Fig. 4j). In addition, most members (except miR-92a-1) and miR-mean correlated positively with the level of METTL3 mRNA with statistical significance (P < 0.05 for all, Fig. 4j, k).
To verify whether METTL3 exerts the onco-promoting role by facilitating pri-miR-17-92 maturation, we constructed a miniMIR17HG plasmid to force expression of 6 miRNAs for rescue studies 14 . This construct contains all miRNAs from the miR-17-92 cluster but no flanking sequences and can be rapidly processed into mature miR-NAs. Transfection of miniMIR17HG in METTL3-knockeddown cells significantly raised all miRNAs (all P < 0.01) to similar levels of control cells (Fig. 4l). In addition, mini-MIR17HG partially rescued cell proliferation (P < 0.05 from day 2, Fig. 4m) and cell migrations (P = 0.0019, Fig. 4n).

METTL3 inhibits PTEN/TMEM127 expression and activates AKT/mTOR pathway by facilitating biogenesis of miR-17-92 cluster
To determine the downstream targets of the miR-17-92 cluster in gastric cancer progression, we performed miRNA cluster enrichment analysis and found the members of the miR-17-92 cluster were significantly enriched in several cancer-related terms, including cell proliferation, AKT pathway, angiogenesis, onco-miR-NAs, and apoptosis (Fig. 6a). Also, we obtained 2056 target genes of the miR-17-92 cluster predicted by miRDB and 525 mRNAs that negatively correlated with METTL3 with a correlation coefficient < −0.18 in the TCGA-STAD database. The two gene pools shared a total of 98 genes, which were enriched in PI3K/mTOR, p53, TGF-β, and MAPK pathways by KEGG analysis (Fig. 6b). Thus, we chose the members that were clustered in the PI3K/mTOR pathway, PTEN and TMEM127, for further study.
In the TCGA-STAD databases, both PTEN and TMEM127 were negatively correlated with METTL3 expression with statistical significance (both P < 0.0001, Fig. 6c). These negative correlations between METTL3 mRNA and PTEN/TMEM127 mRNA were confirmed in our clinical samples by quantitative RT-PCR (Fig. 6d). Nonsynonymous mutation and copy number variations (CNVs) of PTEN occurred in about 10.6% of gastric patients (Fig. 6e). Among them, all CNVs of PTEN were deletion instead of amplification (Fig. 6e). In contrast, nonsynonymous mutations and CNVs of TMEM127 only occurred in 1.6% of the patients, and all CNVs were amplification (Fig. 6e).
By RT-PCR, METTL3 overexpression significantly reduced mRNA levels for PTEN (P < 0.0001) and TMEM127 (P = 0.0019, Fig. 7a), while METTL3 downregulation elevated their mRNA levels (P = 0.0002 and 0.0083, Fig. 7b). Furthermore, forced expression of miniMIR17HG in METTL3-reducing cells reversed (of PTEN, P = 0.0124) or diminished (of TMEM127, P = 0.0929) the differences (Fig. 7b). By western blot and IHC, both PTEN and TMEM127 were remarkably reduced in METTL3-high cells, subcutaneous xenografts, and peritoneal implants (Fig. 7c, g). Accordingly, when METTL3 was reduced, PTEN and TMEM127 were significantly elevated in cells and subcutaneous xenografts (Fig. 7h-j). What is more, the overexpression of (see figure on previous page) Fig. 4 METTL3 promotes tumor progression by facilitating biogenesis of miR-17-92 cluster. a Correlation between METTL3 mRNA and all miRNAs in the TCGA-STAD database. Pearson correlation coefficients were transformed into Z-scores (y-axis) and ranked in descending order (x-axis). Blue and red dots represent negatively and positively METTL3-correlated miRNAs, respectively. Green dots represent the members of the miR-17-92 cluster. b Correlations between METTL3 (y-axis) and members of the miR-17-92 cluster (x-axis). Data were derived from the TCGA-STAD database. PTEN and TMEM127 caused by METTL3-knockdown was reversed by miniMIR17HG (Fig. 7k).

Gastric cancer with METTL3 overexpression is more sensitive to everolimus
Due to the remarkable impact of the METTL3/miR-17-92 cluster on the AKT/mTOR pathway, we explored whether everolimus, an mTOR inhibitor, could inhibit the onco-promoting role of METTL3. Cell viability was measured in cell models with different levels of METTL3 after treatment with solvent or different concentrations of everolimus. Without everolimus, METTL3 overexpression increased cell viability (P < 0.05 from day 1, Fig. 8a), whereas METTL3 downregulation decreased cell viability (P < 0.05 from day 1, Fig. 8b), compared to their corresponding controls. Addition of everolimus remarkably suppressed cell viability, regardless of its concentrations or METTL3 expression (Fig. 8a, b). However, The indicated significances were for comparisons between cells overexpressing pri-miR-17-92-WT and A879C. n = 3 for each group. Data are presented as mean ± SD. ns P > 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.
METTL3-overexpressing cells showed more sensitivity to everolimus in a dose-dependent manner than the control cells (P < 0.05 at day 2 and 3, Fig. 8a). In accordance, the METTL3-low cells showed lower sensitivity to everolimus in both concentrations than the control cells (P < 0.05 at day 2 and 3, Fig. 8b).
With in vivo models, everolimus was effective in reducing tumor volumes in both METTL3-overexpressing or control tumors (Fig. 8c). The tumor inhibition rate of everolimus in METTL3-high tumors was significantly higher than that in control tumors (89.92% vs. 73.26%, P = 0.0465, Fig. 8c-e), suggesting that METTL3-high tumors could be inhibited by everolimus more effectively.

Discussion
Due to its heterogeneity and specificity, gastric cancer has been gradually shown to lack common driven mutations or CNVs that are commonly seen in lung, breast, or colorectal cancer. Consequently, it is difficult to copy the success of the translational therapies based on genetic alterations in certain types of cancers, such as the EGFR or ALK-driven non-small cell lung cancer. Increasing scholars believe that cancer is an epigenetic disease, and disrupted and unstable epigenomes exist widely among different types of tumors with heterogeneity and therapeutic resistance 15,16 . m 6 A is the most abundant internal modification in eukaryotic RNA and may represent a critical mechanism  Table with the KEGG results (pathways and involved genes) of the intersectional genes. c Correlation between METTL3 (y-axis) and PTEN or TMEM127 (x-axis) in the TCGA-STAD database. d Correlation between METTL3 (y-axis) and PTEN or TMEM127 (x-axis) in our clinical cohorts. e Genetic change of PTEN and TMEM127 in gastric cohorts from online databases. Original data were extracted from cBioPortal. FDR, false discovery rate adjusted P value.
for regulating malignant behaviors of tumors 17 . Our data show that the m 6 A level was upregulated in gastric cancers and suppressing m 6 A by METTL3-knockdown hindered tumor growth and metastasis in vitro and in vivo. Based on these discoveries, we believe that abnormal m 6 A modification is an important epigenetic feature of gastric cancer and a potential therapeutic target for further study.
As the predominant "writer" of m 6 A, METTL3 is dysregulated and plays dual roles in cancers, coupled with different substrates and cell types 18 . It directly regulates transcription, translation, and RNA maturation of a broad range of oncogene and onco-suppressors, many of which are undiscovered, and the role of METTL3 cancers depends on orchestration of multiple effects 18 . In non-tumor cells, METTL3 methylated primary miRNAs, such as pri-let-7e and pri-miR221, and facilitated their maturation 19 . Recent studies reported that METTL3 accelerated the maturation of pri-miR221/222 20 and pri-miR-25 21 in bladder and pancreatic cancers. So far, there has been no study reported the regulating function of METTL3 on ncRNA in gastric cancer. As we found, one of the most prominent substrates of METTL3 is a non-coding primary miRNA, pri-miR-17-92, a 6-tandem stem hairpin-containing polycistronic transcript of the gene MIR17HG that encodes a miRNA cluster composed of 6 onco-miRNAs 22 . Through m 6 A modification, METTL3 facilitated pri-miR-17-92 to bind to DGCR8, which recognizes the stem-flanking junctions and orients DROSHA to cleave primary miRNA into precursor miRNA 23,24 . Our study revealed that the over-m 6 A modification on pri-miR-17-92, instead of its overexpression, caused upregulation of the miR-17-92 cluster and gastric cancer progression. To our knowledge, this is the first report of METTL3-mediated-m 6 A-dependent maturation of ncRNA and its downstream signal pathway and biological effect in gastric cancer. Most importantly, we mapped pri-miR-17-92 and found its A879 was the dominant site responsible for the m 6 A-mediated pri-miR-17-92 maturation because A879C mutation significantly abolished the METTL3-mediated m 6 A modification, DGCR8 binding, and miR-17-92 cluster formation. Therefore, the pri-miR-17-92 A879 represented the key mediator of the m 6 Amediated pri-miR-17-92 maturation and was likely a highly precise target for intervention as an epigenetically modifiable position.
The regulation of the miR-17-92 cluster remains largely unclear, other than that c-Myc directly binds to MIR17HG and promotes pri-miR-17-92 transcription 25 . Our findings uncovered another layer of regulation of this miRNA cluster by m 6 A-mediated processing. Widely accepted as an onco-miRNA 26 , the miR-17-92 cluster targets many , and control (lvCTL and shCTL) cells, with (5 or 50 µg/mL) or without everolimus (E). c Tumor volume of mice carrying METTL3-high or control tumors and fed with everolimus or solvent. Mice were pre-inoculated with MKN-45 tumor cells, administrated with everolimus or solvent daily from day 0 to 16, then sacrificed at day 17. d Photos of xenograft tumors when mice were sacrificed. e Tumor inhibition rate of everolimus in METTL3-high and control tumors. Tumor inhibition rate = 1 − the tumor weight with everolimus/the corresponding tumor weight with solvent. f Overview of METTL3/m 6 A-mediated miRNA cluster biogenesis and AKT/mTOR pathway activation in gastric cancer development. n = 3 for each group. Data are presented as mean ± SD. *P < 0.05. onco-suppressing genes, such as SMAD2 and SMAD4 27 , p21 28 , and TRAF3 29 . In this study, we found that TMEM127 and PTEN were part of the targets. Furthermore, forced expression of the miR-17-92 cluster counteracted the effects of METTL3-knockdown in PTEN/ TMEM127 regulation, cell proliferation, and metastasis. TMEM127 is a negative regulator of the mTOR pathway and a tumor suppressor located on chromosome 2q11 30 . So far, the function of TMEM127 in gastric cancer and its association with the miR-17-92 cluster have not been reported. In this study, we did not observe frequent mutations or deletions of TMEM127. Instead, TMEM127, along with PTEN and the mTOR pathway, was efficiently regulated by the miR-17-92 cluster as its target. Therefore, we suspect that METTL3/miR-17-92 cluster activates mTOR pathways by targeting PTEN and TMEM127 in gastric cancer progression. Of note, several other signaling pathways, including p53, TGF-beta, and MAPK signaling pathways, were probably also involved, which are of value for further investigation in future studies.
Currently, drugs targeting epigenetic changes, such as DNA methyltransferase inhibitors decitabine, have become standard treatments in hematological malignancies 31 . Up to now, the small molecule inhibitor against METTL3 is not available. Based on the above discovery, the mTOR inhibitor everolimus was chosen to interfere with METTL3/miR-17-92 cluster/TMEM127 or PTEN/ mTOR signaling pathway in gastric cancer. We found that everolimus indeed reversed the METTL3-induced proliferation in a dose-dependent manner. This effect was remarkably pronounced when METTL3 was highly expressed. The high sensitivity to everolimus in METTL3high cells could be because the mTOR pathway was greatly activated by METTL3 in these cells. These data confirmed that the oncogenic role of METTL3 relies on mTOR activation in another aspect. Everolimus has shown promising efficacy in patients with previously treated advanced gastric cancer in a phase II study 32 . However, phase III studies [33][34][35] failed to repeat the positive results, except for that the progression-free survival was significantly prolonged. Our findings provided evidence that the METTL3 level might be a potential predictor for treatment efficacy of everolimus in gastric cancer and may help screen possible advantageous subgroups out from the previous failed clinical trials with everolimus-treated gastric cancer. These inferences are yet to be further studied and confirmed in clinical trials.
In conclusion, METTL3/m 6 A promotes gastric cancer growth and metastasis by facilitating pri-miR-17-92 processing into the oncogenic miRNA cluster and activating the AKT/mTOR pathway by targeting PTEN and TMEM127, which could be targeted by everolimus (Fig. 8f). These findings provide us with a novel insight into the role of METTL3 in the regulation of cancer development and a theoretical rationale for use of everolimus in the treatment of m 6 A/METTL3-high gastric cancer.