Identification and validation of crucial lnc-TRIM28-14 and hub genes promoting gastric cancer peritoneal metastasis

Abstract

Background: Gastric cancer peritoneal metastasis (GCPM) is an important cause of cancer-related deaths worldwide. Long non-coding RNAs (lncRNAs) play a key role in the regulation of GCPM, but the underlying mechanisms have not been elucidated.

Methods: High-throughput RNA sequencing (RNA-seq) was performed on four groups of clinical specimens (non-metastatic gastric cancer primary tumor, adjacent normal gastric mucosal tissue, gastric cancer primary tumor with peritoneal metastasis and adjacent normal gastric mucosal tissue). After sequencing, many lncRNAs and mRNAs were screened for further Weighted Gene Co-expression Network Analysis (WGCNA). GCPM-related hub lncRNAs and genes were identified by cytoHubba and validated by Quantitative real-time PCR (qRT-PCR), Receiver operating characteristic curve (ROC) analysis and Kaplan-Meier survival analysis. GO, KEGG and GSEA showed GCPM-related pathways. Correlation analysis revealed the potential relationship between hub lncRNAs and genes.

Results: By analyzing lncRNA expression data by WGCNA, we found that blue module was highly correlated with GCPM (r = 0.44, p = 0.04) and six lncRNAs involved in this module (DNM3OS, lnc-MFAP2-53, lnc-PPIAL4C-4, lnc-RFNG-1, lnc-TRIM28-14 and lnc-YARS2-4) were identified. We then performed qRT-PCR validation of gastric cancer specimens and found that the expression of lnc-RFNG-1 and lnc-TRIM28-14 was significantly increased in gastric cancer tissues with peritoneal metastasis. Kaplan-Meier survival analysis showed shorter overall survival time (OS) for gastric cancer patients with high expression of lnc-TRIM28-14. Receiver operating characteristic curve (ROC) analysis showed that lnc-TRIM28-14 could improve the sensitivity and specificity of GCPM diagnosis. In addition, we identified three key mRNAs (CD93, COL3A1 and COL4A1) associated with gastric cancer peritoneal metastasis through WGCNA analysis and clinical specimen validation. Moreover, there was a positive correlation between lnc-TRIM28-14 and the expression of CD93 and COL4A1 in gastric cancer peritoneal metastasis, suggesting a regulatory relationship between them. Subsequent GO, KEGG and GSEA analysis suggested that ECM-receptor interaction and focal adhesion were the hub pathways of GCPM.

Conclusion: In summary, lnc-RFNG-1, lnc-TRIM28-14, CD93, COL3A1 and COL4A1 could be novel tumor biomarkers and potential therapeutic targets for GCPM.

Keywords: Gastric cancer; Peritoneal metastasis; RNA-seq; WGCNA; lncRNA.

Fig. 1

Expression profile of lncRNAs in gastric cancer tissues and adjacent tissues. A RNA sequencing and analysis flowchart. B Boxplot of FPKM values of lncRNAs in each sample. C Principal component analysis based on the expression of lncRNAs. D Novel lncRNAs classification. E–G Heatmap showing lncRNAs expression in PTvsPA E, MTvsMA F, MTvsPT G. H-J Volcano plot showing the lncRNAs expression pattern in PTvsPA H, MTvsMA I, MTvsPT J

Fig. 2

Construction of lncRNA co-expression clusters. A The sample clustering tree. B Variation of scale independence of co-expression networks under different soft thresholds. C Changes in mean connectivity of co-expression networks under different soft thresholds. D The hierarchical cluster dendrogram showing 12 co-expression lncRNA modules

Fig. 3

The correlations between lncRNA modules and GCPM. A Topological overlap matrix heatmap with one lncRNA per row and column. B Heatmap showing the lncRNA module-GCPM correlation. C Scatter plot showing lncRNA-trait significance (correlation between lncRNA expression and GCPM) and module membership (correlation between lncRNA expression and module eigengenes) in blue module. D The co-expression relationship between lncRNAs in blue modules was visualized using Cytoscape. E The top 15 lncRNAs with highest degree Calculated by cytoHubba

Fig. 4

Tissue validation of candidate lncRNAs. A-F The expression levels of lncRNAs in gastric cancer tissue as determined by qRT-PCR. Student's t test, *, p < 0.05, **, p < 0.01, ***, p < 0.001, ****, p < 0.0001, ns, not significant. G Representative RT-PCR plots of the candidate lncRNAs. PCR products were separated by 2% agarose gel electrophoresis under 100 V for 1 h. Images were processed by Adobe Photoshop 22.1.1 software. GC-NPM, Gastric cancer patients without peritoneal metastasis; GCPM, Gastric cancer patients with peritoneal metastasis; N, tissue adjacent to tumor; T, tumor tissue. H-M ROC curve of six lncRNAs signature for the detection of GCPM. AUC, area under curve, Sen, sensitivity at the optimal cut-off value, Spe, specificity at the optimal cut-off value. N-S Kaplan-Meier survival analysis of overall survival (OS) for the lncRNA-high-expressed group and lncRNA-low-expressed group using log-rank test

Fig. 5

Construction of gene co-expression clusters. A The sample clustering tree. B Variation of scale independence of co-expression networks under different soft thresholds. C Changes in mean connectivity of co-expression networks under different soft thresholds. D The hierarchical cluster dendrogram shows 18 co-expression modules

Fig. 6

The correlation between gene modules and GCPM. A Topological overlap matrix heatmap with one gene per row and column. B Heatmap showing the correlation between gene modules and GCPM. C Scatter plot showing gene-trait significance (correlation between gene expression and GCPM) and module membership (correlation between gene expression and module eigengenes) in magenta module. D The co-expression relationship between genes in blue modules was visualized using Cytoscape. E Top 15 genes with highest degree calculated by cytoHubba

Fig. 7

Tissue validation of candidate genes. A-F Expression level of candidate hub genes expression in gastric cancer tissue by qRT-PCR. Student's t test, *, p < 0.05, **, p < 0.01, ***, p < 0.001, ****, p < 0.0001, ns, not significant. G-L ROC curve of six genes signature used to determine GCPM. AUC, area under curve, Sen, sensitivity at the optimal cut-off value, Spe, specificity at the optimal cut-off value. M-R Kaplan-Meier analysis of the OS for the gene-high-expressed group and gene-low-expressed group using log-rank test

Fig. 8

GCPM-related pathway analysis. A-B. GO terms and KEGG pathways enriched by megenta module genes. C-D GO terms and KEGG pathways enriched by lnc-TRIM28-14 co-expressed genes. E Venn plot for KEGG terms between B and D. F Venn plot for pathways associated with the expression of lnc-TRIM28-14 by GSEA and overlapping pathways in E. G-H GSEA analysis showed that high expression of lnc-TRIM-28–14 was significantly associated with ECM-receptor interaction and focal adhesion

Fig. 9

Identification of the relationship between hub lncRNAs and genes. A Correlation heatmap of lncRNAs and genes determined by RNA-seq with the threshold of p < 0.05. B-G Expression correlation analysis for lnc-TRIM28-14 and 6 candidate hub genes determined by qRT-PCR (normalized to GAPDH). H-M Expression correlation analysis for lnc-RFNG-1 and 6 candidate hub genes based on qRT-PCR assay (normalized to GAPDH)