We used the human protein atlas (HPA, https://www.proteinatlas.org/) to examine TPM4 mRNA expression in normal tissues (N= 13,084) across Genotype-Tissue Expression (GTEx https://gtexportal.org/) (16). TPM4 mRNA expression in normal and tumor tissues (N= 15,776), the expression of TPM4 within tumors as well as within paired normal tissues (n=15043) and data from UCSC XENA (https://xenabrowser.net/datapages/) were used. We used GEPIA2’s “Stage Plot” module to evaluate the correlation between TPM4 expression, which was upregulated, and the pathological stages of cancers. To perform the analysis and comparison, RNAseq data from TCGA and GTEx were processed uniformly using the Toil process (17) and then log2-transformed. Xiantao tool (https://www.xiantao.love/ is a useful bioinformatics analysis web tool, and was used for visualization. Statistical analysis was performed using the Wilcoxon rank-sum test, and significant outcomes were defined at p < 0.05. We used the immunofluorescence staining images of three human cancer cell lines (A-431, U251MG, and U-2 OS) to display the subcellular localization of TPM4 in cancer cells from the HPA dataset.

We assessed variations within TPM4 expression for the diagnosis and prognosis of cancer using RNA sequencing data obtained from TCGA (https://portal.gdc.cancer.gov/). Xiantao tool was analyzed statistically using the log-rank test, with a P-value < 0.05 regarded as significant. We obtained the ''hazard ratio (HR) 95% CI'' as well as the ''P-value'' and used Xiantao tool to visualize the forest plot. The receiver operating characteristic (ROC) curve and area under the ROC curve were applied to evaluate the diagnostic value of TPM4 in pan-cancer tissues using Xiantao tool . Diagnostic value: low accuracy (AUC: 0.5–0.7), certain accuracy (AUC: 0.7–0.9), and high accuracy (AUC > 0.9).

cBioPortal (18) (http://www.cbioportal.org) provides a platform for analyzing and interpreting cancer genetic data and facilitates the interpretation of molecular data acquired from cancer histological and cytological studies. Gene alteration data from 2683 samples collected from 2565 pan-cancer patients obtained from UCSC Xena and the International Cancer Genome Consortium(ICGC) (https://www.icgc-argo.org) data portal from "TCGA pan-cancer Atlas Studies" were used for analysis. The mutation landscape of TPM4, including the mutation type, copy number alteration(CAN), as well as mutation frequency data, was searched using the module titled "Cancer Types Summary". Somatic mutation datasets from the publicly available TCGA database were acquired via the data portal to identify the genomic data commons (https://portal.gdc.cancer.gov/). The dataset includes data from patients with stomach adenocarcinoma (STAD) within the TCGA database with high TPM4 expression (n=212) and low TPM4 expression (n=21); these data were used with HOME for research, a useful online bioinformatic tool (https://www.home-for-researchers.com).

UALCAN (http://ualcan.path.uab.edu/analysis.html) (19) was used to explore the promoter DNA methylation levels in TPM4 in normal and pan-cancer tissues. The beta value represents the DNA methylation level, hypomethylation (beta: 0.3–0.25), and hypermethylation (beta: 0.7–0.5) (20). The DNA methylation map of TPM4 in STAD was obtained from the MethSurv database (21) ("Gene visualization" module). mRNA modification analysis of 45 methylation regulators involved in n1-methyladenosine (m1A) and 5-methylcytosine (m5C) n6-methyladenosine (m6A) addition in pan-cancer tissues across TCGA was performed using SangerBox3.0, a helpful bio information online tool ("pan-cancer analysis-Mrna modification" module) (http://sangerbox.com/).

The European Prospective Investigation into Cancer and Nutrition (EPIC) (22) (https://gfellerlab.shinyapps.io/EPIC_1-1/) is an online platform that provides the infiltration ratio of eight types of immune cells according to the expression information Scores for stromal, immune, and ESTIMATE cells within the TCGA. The STAD dataset was derived using the ''estimate'' (23) R package, which is used to estimate the tumor purity scores. TIMR2.0 (24) (http://timer.comp-genomics.org/) serves as a platform for analyzing the immunological characteristics of cancer in a systematic manner across TCGA. To determine the association with TPM4 expression as well as eight immune checkpoints, a module named "Gene_Corr" was employed. After we extracted the "P-value" and "r-value," we used the Xiantao tool to visualize using a heatmap. Relationships between TPM4 expression and the MSI, TMB, and NEO were obtained using SangerBox3.0 ("pan-cancer analysis - heterogeneity analysis" module).

Five online prediction databases for miRNAs, namely RNA22 (http://cbcsrv.watson.ibm.com/rna22.html/) (25), DIANA-mircoT (http://diana.imis.athena-innovation.gr/DianaTools/index.php?r=microT_CDS/index) (26), miRcode (http://www.mircode.org/index.php) (27), miRWalk (http://mirwalk.umm.uni-heidelberg.de/) (28), and miRDB (http://mirdb.org/miRDB/) (29) were used for predicting the TPM4 target miRNAs. miRNAs that were retrieved in at least three databases were defined as target miRNAs. StarBase2.0 (https://starbase.sysu.edu.cn/) was used for analyzing the lncRNA-mirRAN interactome and the relevance of miRNA and TPM4 (lncRNA). ''Mammals, humans, hg19, strict stringency (≥5) of CLIP-Data, including or excluding Degradome-Data'' was used as the screening criteria. A Sankey diagram of the miRNA-lncRNA interactome and lncnRNA-miRNA-TPM4 network was visualized using the Xiantao tool and Cytoscape, respectively.

GSCALite (http://bioinfo.life.hust.edu.cn/web/GSCALite/) (30) is used for analyzing integrated mRNA expression, mutation, immune infiltration, methylation across TCGA datasets, and drug resistance datasets from GDSC (https://www.cancerrxgene.org/), and CRTP (http://portals.broadinstitute.org/ctrp/). We analyzed the drug sensitivity of the TPM family (TPM1, TPM2, TPM3, and TPM4) using data from GDSA and CRTP. According to the retrieved datasets, Food and Drug Administration (FDA)-approved chemotherapy drugs related to TPM4 expression were found in drug banks (https://go.drugbank.com/drugs). The outcomes were visualized with Xiantao tool ." Connectivity Map (CMap) (31) is an expression profiling database based on the expression of intervening genes to reveal functional links between small molecule compounds, genes, and disease states. Xiantao tool was used to identify the top 100 up-regulated and down-regulated differentially expressed genes (DEGs) between the TPM4 high-expression and low-expression groups. These DEGs were applied to querying against CMap to predict small molecular potential therapeutic drugs for cancer patients. Drugs with positive/ negative connectivity scores can induce/reverse effects against the input signature in human cell lines.

The top 300 genes showing positive relative co-expression and top 300 genes showing negative relative co-expression with TPM4 (|cor|>0.3, P below 0.05) were identified from TCGA using Xiantao tool for visualizing the heatmaps. The PPI network of the top 100 genes showing positive relative co-expression with TPM4 was identified using STRING (https://cn.string-db.org/), which lists publicly available PPI data (32). Hub genes were analyzed using ''MOCODE'' and ''CytoHubba'' in Cytoscape (edition 3.7.2). Enrichment analyses based on Gene Ontology and Kyoto Encyclopedia of Genes and Genomes were conducted for the top 300 genes showing positive relative co-expression with TPM4 using the Xiantao tool " intended for cluster information analysis.

AGS and BGC-823 human GC cells were purchased from the Shanghai Cell Bank of the Chinese Academy of Sciences (Shanghai, China), cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum (FBS) (Gibco, Grand Island, NY, USA) and 1% penicillin-streptomycin, and maintained at 37°C and 5% carbon dioxide. Recombinant lentiviral vectors for TPM4 RNAi (LV-TPM4) and lentiviral vectors for negative control (LV-Ctrl) were designed and packaged in 293T cells from GeneChem Co., Ltd. The TPM4 siRNA had the following sequence: 5′-GGAGGACAAATATGAAGAAGA-3′. The shTPM4/shCtrl cohorts had AGS/BGC-823 cells (5 × 103/well) subcultured in 96-well culture plates and infected with LV-TPM4/LV-Ctrl. The cells that were infected were selected by incubation with 2 µg/mL puromycin for 48 h. The efficiency for TPM4 knockdown was detected by western blotting.

We performed RT-PCR to determine TPM4 mRNA expression in GC cells as well as the knockdown efficiency of the TPM4 TRIzol® Plus Purification Kit for RNA (12183-555; Invitrogen; Thermo Fisher Scientifc, Inc.). We referred to the kit instructions for generating reverse transcription cDNA and detecting PCR products in a fluorescent quantitative PCR instrument. GAPDH (glyceraldehyde-3-phosphate dehydrogenase) was used to be an internal control. The oligonucleotide primers indicated as follows were used for quantitative PCR: TPM4, 5'- TTGAGGAGGAGTTGGACAGGG-3' forward and 5'-CCAGGATGACCAGCTTACGAG-3' reverse; GAPDH, 5’-TGACTTCAACAGCGACACCCA-3' forward and 5’-CACCCTGTTGCTGTAGCCAAA-3’ reverse. The reaction conditions were as follows: 45 cycles, 95°C: pre-change for 15 s, 95°C: denaturation for 5 s, 60°C: annealing and extension for 30 s. The 2-ΔΔCt method was used to analyze the expression levels of each gene.

After the cells were digested with protein lysates, the total proteins of the AGS and BGC-823 cells were extracted. A BCA kit (Beyotime, P0010) was used for measuring the cellular protein content. 10% SDS-PAGE was used to separate different proteins, and 50 g of protein was loaded per lane. The proteins were transferred to polyvinylidene difluoride (PVDF) membranes and blocked with 5% milk. The primary rabbit antibodies used were anti-TPM4 (cat number: ab181085; 1:1,000; Abcam) and anti-GAPDH (cat number: sc-32233; 1:1,000; Santa Cruz). A processing film containing HRP-conjugated goat anti-rabbit IgG (cat. no. sc-2004; 1,000; Santa Cruz) and anti-mouse IgG (cat. no. sc-2005; 1,000; Santa Cruz) was used. The membranes were detected using an enhanced chemiluminescence detection system (Pierce; Thermo Fisher Scientific, Inc.) and visualized using the ChemiDoc system (BioRad Laboratories, Inc.). The intensity of the proteins was measured using ImageJ (edition 1.8.0; National Institutes of Health, Bethesda, MD, US).

AGS and BGC-823 cells (density 2.0 ×10⁵ cells/well) transfected with LV-TPM4/LV-Ctrl were subcultured in 6-well plates (37°C, 5% CO₂ in an incubator), in a culture system of 100 μL/well for 24 h. The cells within the plate were scratched using a scratch tool. The serum-free medium was substituted, and images were acquired under a microscope (XDS-100, Cai Kang Optical Instrument Co, Ltd, China) at 0, 8, and 24 h.

Transwell kits (cat.NO 2433 Corning, US) were applied. AGS and BGC-823 cells transfected with shTPM4 and shctrl were planted into the upper chamber (8 µm) at a density of 0.8 × 105 cells/well in a serum-free medium. For the transwell assay, the medium within the upper chamber was removed, 100 μL of serum-free medium was added, and 600 μL of 30% FBS medium was added for 16 h at 37°C. For the Matrigel transwell assay, 100 μL of serum-free medium was applied to the upper and lower chambers. A layer of matrigel matrix glue (Corning) (ratio of serum-free medium: matrix glue = 8:1) was coated within the lower chamber, and the cells were incubated for 24 h at 37 °C. Non-metastatic cells were removed from the chamber. The chamber was fixed with a 4% paraformaldehyde fixative for 30 min. Later, the cells were transferred by staining with 1% crystal solution on the membrane's lower surface for 1-3 min. The cells on the lower side of the membrane were counted. Images were recorded under a microscope (BX53, Olympus Company, Japan).

Statistical analyses in this study were conducted using the above online database and ''R package (R studio edition: 1.2.1335, R edition: 3.6.3), as described above. GraphPad Prism 9.0 (GraphPad Software La Jolla, CA, USA) was used for the statistical analysis of experimental data. Differences were compared using a Student's t-test, and outcomes are shown as mean ± SD. Statistical significance was reported at *P<0.05, **P< 0.01, ***P <0.001, and ****P < 0.0001.

The HPA datasets showed that the top-ranked expression of TPM4 was within the urogenital, respiratory, and digestive systems ( Figure 2A and Supplementary Table 1 ). The TCGA and GTEx datasets revealed that TPM4 was significantly upregulated (P < 0.05) in BRCA, CHOL, COAD, DLBC, ESCA, GBM, HNSC, LAML, LGG, LIHC, LUCS, OV, PAAD, PCPG, PEAD, READ, SARC, STAD, and TGCT. In contrast, TPM4 was downregulated (P < 0.05) in BLCA, KICH, KIRP, LUAD, PRAD, LUSC, SKCM, THCA, THYM, UCEC, and UCS ( Figure 2B and Supplementary Table 2 ). GEPIA2's dataset indicated PAAD and SKCM were significantly associated with tumor stage (P<0.05, Supplementary Figure 1 ). According to the outcomes of the TCGA dataset analysis, TPM4 was differentially expressed in tumor tissues and paired normal tissues in BLCA, BRAC, CHOL, COAD, ESCA, HNSC, KICH, KIRC, KIRP, LIHC, LUAD, PRAD, STAD, and UCEC ( Figure 2C and Supplementary Table 3 ). The subcellular localization of TPM4 to actin filaments ( Figure 2D ) and the cytosol ( Figures 2E, F ) was observed. Therefore, our outcomes revealed that TPM4 expression was upregulated in most tumors and was greater than that in paired (unpaired) normal tissues in STAD.

Based on the Forest plot and the findings of Cox assessment ( Figure 3A ), TPM4 is an adverse factor for overall survival (OS) in ACC, CESC, CHOL, DLBC, ESCC, HNSC, KIRC, KIRP, LIHC, LUAD, LUSC, MESO, PAAD, STAD, UCEC, and UVM (P < 0.05, HR > 1), whereas it acts as a potentially beneficial factor in COAD, COADREAD, OS, and PCPG (P < 0.05, HR < 1). Furthermore, we focused on the association involving TPM4 expression and digestive system malignancies, with a high expression of TPM4 linked significantly to poor prognosis in, LIHC ( Figure 3D , p=0. 02), ESCC ( Figure 3E , p=0.028), PAAD ( Figure 3F , p < 0.001), and STAD ( Figure 3G , p=0.018). Notably, patients with low TPM4 expression show worse prognosis than patients with high TPM4 expression in cases of COAD ( Figure 3B , p=0.04) and COADREAD ( Figure 3C , p=0.009). Next, we used ROC curves to assess the diagnostic efficacy of TPM4 in digestive cancers. TPM4 had a certain accuracy in predicting COAD (AUC = 0.807) ( Figure 3H ), COADREAD (AUC = 0.837) ( Figure 3I ), ESCA (AUC = 0.725) ( Figure 3J ), LICH (AUC = 0.739) ( Figure 3K ), and STAD (AUC = 0.795) ( Figure 3L ). TPM4 expression also had a high accuracy in predicting PAAD (AUC = 0.972) ( Figure 3M ). Collectively, TPM4 expression has diagnostic and prognostic value in different cancers, including STAD.

We investigated the pan-cancer genetic alterations in TPM4 across the cBioPortal. As shown in Figure 4A , TPM4 expression was altered in 131 samples collected from 2565 patients with different cancer types, which accounted for 5% of the samples. The TPM4 mutation rate was observed in 32 types of cancer, as shown in Figure 4B . The highest mutation rate (>60%) was observed in cases with OV. Higher mutation rates were not observed in 24 cancers, including STAD. The most common alterations observed were the "mutation" and "amplification" types in copy number variation (CNV) in STAD. Following this, as shown in Figure 4C , we investigated the correlation between putative copy-number alteration (CNA) in TPM4 and TPM4 mRNA expression in pan-cancer tissues. Fifteen mutated genes were identified within the mutation spectrum of TPM4 high/low expression cohorts in STAD. The top 5 genes were ABCA12, DOCK3, NALCN, PCDH17, and KRAS ( Figure 4D ).

We assessed the promoter DNA methylation levels in TPM4 in normal tissues and 19 types of cancer tissues. TPM4 was hypermethylated in BRCA, CHOL COAD, ESCA, HNSC, LUAD, PCPG, PRAD, THYM, and UCEC. On the contrary, it was hypomethylated in GBM, KIRC, LICH, PAAD, SARC, STAD, TGCT, and THCA ( Figure 5A ). Following this, according to the methylation map of TPM4 from the MethSurv database, 25 CpG sites of TPM4 were identified in STAD ( Figure 5B and Supplementary Table 4 ). In addition, the mRNA modification parameter is a crucial component of epigenetics and is involved in post-transcriptional gene regulation. Many studies have shown that mRNA modification is closely related to cancer progression and incidence (33). m1A, m5C, and m6A are common types of mRNA modification. To explore the correlation between TPM4 expression and 45 mRNA modification regulators (see Supplementary Table 5 ), methyltransferases (writers), demethylases (erasers), and RNA-binding proteins (readers) were selected. As shown in Figures 6A–C , TPM4 expression was positively related to most m1A, m5C, and m6A methylations in pan-cancer tissues. Subsequently, in STAD, the top 10 methylation regulatory factors included METTL14 (m6A_writer; r=0.403, P < 0.0001), YTHDF3 (m1A_reader; r=0.421, P < 0.0001), FTO (m6A_eraser; r=0.399), TET2 (m5C_eraser; r=0.375), NSUN3 (m5C_writer; r=0.374), WTAP (m6A_writer; r= 0.368), YTHDC2 (m6A_reader; r=0.366), CBLL1 (m6A_reader; r=0.357), FMR1 (m6A_reader; r= 0.338), and YTHDC1 (m1A_reader; r= 0.306). The above outcomes suggest that TPM4 expression is closely associated with DNA methylation and mRNA modification in different cancers, including STAD.

We investigated the association between TPM4 expression and immune cell infiltration within the tumor microenvironment (TME). The EPIC online tool showed that eight cancer-associated immune cells were related to TPM4 expression in different cancers, especially in KIPR, LUAD, PCPG, PRAG, STAD, and UVM ( Figure 7A ). The ESTIMATE score is useful for determining tumor purity and immune cell infiltration within the TME. Our findings revealed that TPM4 expression is positively related to the ESTIMATE score in STAD ( Figure 7B ). Next, we assessed the enrichment scores of TPM4 high- and low-expression cohorts in immune cells, including CD8 T+ cells, eosinophils, macrophages, NK, and Treg cells in STAD. The enrichment scores within the two cohorts showed significant differences ( Figure 7C ). The immune checkpoint (ICP) gene was found to play a role in immune cell infiltration and immunotherapy outcomes (34). Our result indicates that TPM4 expression was positively related to the expression of these eight ICP genes (CD274, CTLA4, HAVCR2, LAG3, PDCD1, PDCD1, LG2, and TIGIT) in BRCA, CHOL, COAD, DLBC, ESCA, KICH, LGG, LIHC, LUAD, PAAD, PRAD, READ, STAD, THCA, and UVM ( Figure 7D and Supplementary Table 6 ). This suggests that TPM4 coordinates ICP gene activity through different signaling pathways and may be a pivotal target for immunotherapy. TMB, MSI, and NEO are considered predictors for response to tumor immunotherapy within the TME (35–37). Moreover, we observed the increased expression of TPM4 and the consequent increase within the TMB in ACC, UCSC, GBM, PAAD, and STAD ( Figure 7E and Supplementary Table 7 ). MSI showed a positive association with TPM4 expression in TGCT, SARC, ACC, UVM, and STAD ( Figure 7F and Supplementary Table 8 ). TPM4 expression was positively associated with NEO in ACC, TGTC, DLBC, PCPG, and THCA ( Figure 7G and Supplementary Table 9 ). Collectively, TPM4 may affect antitumor immunity through its association with immune infiltrating cells, ICPS, MSI, TMB, and NEO in pan-cancer tissues.

RNA22, DIANA-micro, miRWalk, miRcode, and TargetScan were used to identify the target miRNAs of TPM4 in STAD, as shown in Figure 8A and Supplementary Table 10 . We identified 7, 24, 2063,10, and 1230 TPM4 target miRNAs from these sources, respectively. Forty-one common miRNAs were predicted in three databases. To further explore the target lncRNAs of 41 miRNAs, Starbase2.0 was used, and seven miRNAs (hsa -miR-613, hsa -miR-338-3P, hsa -miR-206, hsa -miR-30e-5P, hsa -miR-30b-5p, hsa -miR-10b-5p, and hsa -miR-299-3p) showed target lncRNAs. mRNA expression is generally negatively correlated with miRNA expression (38). TPM4 expression was negatively correlated with hsa-miR-30e-5P ( Figure 8B ), hsa -miR-30b-5P ( Figure 8C ), hsa s-miR-338-3 ( Figure 8D ), and hsa -miR-206 expression ( Figure 8E ). lncRNAs could act as competitive endogenous miRNAs to further regulate mRNA expression (39). We also identified four target miRNAs and their target lncRNAs that were regulated in a negative manner ( Figure 8F and Supplementary Table 11 ). Finally, we constructed a lncRNAs-miRNAs-TPM4 regulatory network for GC ( Figure 8G ).

The CTPR dataset indicated the correlation between members of the TPM family (TPM1, TPM2, TPM3, TPM4) mRNA expression levels and drug sensitivity; the top three drugs that were positively related to TPM4 expression were COL-3 (incyclinide), CR-1-31-B (eIF4A inhibitor), and GSK525762A (Bet inhibitor); ( Figure 9A and Supplementary Table 12 ; P < 0.0001). Based on GDSC drug sensitivity outcomes, the top three drugs that were positively related to TPM4 expression were 5-fluorouracil, AR-429 (histone deacetylase inhibitor), and AT-7519 (inhibitor of cyclin-dependent kinases), and the ones that were negatively related to TPM4 expression were 17-AAG (inhibitor of heat shock protein 90), bleomycin (50 uM), CHIR -99021 (GSK-3 α/β inhibitor), docetaxel ( Figure 9C and Supplementary Table 14 , P < 0.0001). As some drugs that feature within the prediction outcomes for CTRP and GDSC drug sensitivity are used in scientific research, 23 ( Figure 9B and Supplementary Table 13 ) and 13 ( Figure 9D and Supplementary Table 15 ) types of TPM4-related antitumor drugs approved by the FDA are based on data from drug banks. We analyzed DEGs with high and low expression of TPM4 by CMap “qury” online tool. Based on the results of CMap database inquiry, 15 types of small molecules drugs including ALK/ BCR-ABL/ BTK/ CDK /Met inhibitor, etc. were identified ( Supplementary Table 16 ), eight small molecular drugs targeting TPM4 obtained ( Supplementary Table 16 ), meaning that they have the potential to treat PRAD. UVM, LUAD, KIRC, COAD, BRCA, HCC. Notably, Rucaparib had the highest absolute value score, meaning that the drug has the potential to treat the7 types of cancer ( Figure 9E ). Rucaparib (40) is a poly (ADP-ribose) polymerase (PARP) inhibitor used to treat recurrent ovarian and prostate cancers.

We identified genes exhibiting co-expression from TCGA. The heatmap shows the top 50 co-expressed genes with their expression positively and negatively correlated with TPM4 expression in STAD ( Figures 10A, B ; Supplementary Table 17 ). The top 100 co-expressed genes with their expression positively related to TPM4 expression are shown in the PPI network ( Figure 10C ; Supplementary Table 17 ). The top 10 hub genes were COL1A2, COL1A1, CLO3A1, COL5A, POSTN, FN1, MMP2, LUM, SPARC, and DCN ( Figure 10D ). The top 5 hub genes were COL1A2, CLO3A1, FN1, MMP2, LUM, SPARC, and DCN ( Figure 10E ). Following this, GO and KEGG enrichment analyses were conducted using the top 300 co-expressed genes. The GO analysis involved molecular function, cellular components, and biological processes. (See Figure 10F and Table 1 ). KEGG pathway enrichment (See Figure 10G and Table 2 ). The outcomes revealed that TPM4 expression plays a role in GC by regulating the extracellular matrix.

To verify the function of TPM4 in GC, we used two types of GC cells in our laboratory. The data showed the mRNA ( Figure 11A ) and protein expression ( Figure 11B ) levels of TPM4. In our previous study, after shRNA lentivirus infection, the knockdown efficiency of shTPM4 was 71.1%, and the knockdown of TPM4 inhibited GC cell proliferation (15) ( Supplementary Figure 2 ). In this research, wound healing and transwell assays were used to determine the migration potential of GC cells. The knockdown of TPM4 repressed the migration of AGS ( Figure 11C , P < 0.01) and BGC-823 ( Figure 11D , P< 0.01), as observed in the wound healing assay. We also found that TPM4 expression inhibited the migration of AGS cells ( Figure 11E , P < 0.01) and BGC-823 cells ( Figure 11F , P < 0.01). The Matrigel transwell assay showed that the invasive potential was significantly reduced in AGS cells ( Figure 11G , P < 0.01) (G) and BGC-823 cells ( Figure 11H , < 0.01) with TPM4 knockdown. The outcomes of these experiments showed that TPM4 is a key oncogene that promotes tumor invasion and cell migration in GC.