Identification and development of a novel invasion-related gene signature for prognosis prediction in colon adenocarcinoma

Abstract

The overall survival of metastatic colon adenocarcinoma (COAD) remains poor, so it is important to explore the mechanisms of metastasis and invasion. This study aimed to identify invasion-related genetic markers for prognosis prediction in patients with COAD. Three molecular subtypes (C1, C2, and C3) were obtained based on 97 metastasis-related genes in 365 COAD samples from The Cancer Genome Atlas (TCGA). A total of 983 differentially expressed genes (DEGs) were identified among the different subtypes by using the limma package. A 6-gene signature (ITLN1, HOXD9, TSPAN11, GPRC5B, TIMP1, and CXCL13) was constructed via Lasso-Cox analysis. The signature showed strong robustness and could be used in the training, testing, and external validation (GSE17537) cohorts with stable predictive efficiency. Compared with other published signatures, our model showed better performance in predicting outcomes. Pan-cancer expression analysis results showed that ITLN1, TSPAN11, CXCL13, and GPRC5B were downregulated and TIMP1 was upregulated in most tumor samples, including COAD, which was consistent with the results of the TCGA and GEO cohorts. Western blot analysis and immunohistochemistry were performed to validate protein expression. Tumor immune infiltration analysis results showed that TSPAN11, GPRC5B, TIMP1, and CXCL13 protein levels were significantly positively correlated with CD4+ T cells, macrophages, neutrophils, and dendritic cells. Further, the TIMP1 and CXCL13 proteins were significantly related to the tumor immune infiltration of CD8+ T cells. We recommend using our signature as a molecular prognostic classifier to assess the prognostic risk of patients with COAD.

Keywords: Colon carcinoma; Gene signature; Invasion-related genes; Molecular subtyping; Prognosis.

Fig. 1

The protocol of colon carcinoma invasion-related prognosis features

Fig. 2

a Samples cluster heatmap with consistent cluster k = 3; b cluster heatmap of 8 prognostic genes; c OS survival curves of all TCGA colon carcinoma samples molecular subtypes; d PFS survival curves of all TCGA colon carcinoma samples molecular subtypes

Fig. 3

a Volcano map of C1 and C3 subtypes DEGs; b Volcano map of C1 and C2 subtypes DEGs; c Volcano map of C2 and C3 subtypes DEGs; d The significant top 10 GO functional annotations in biological process; e The significant top 10 GO functional annotations in cellular component; f M The significant top 10 GO functional annotations in molecular function; g the top 10 KEGG pathway of DEGs

Fig. 4

a Comparison of ssGSEA Immune score among molecular subtypes in TCGA dataset; b Comparison of Mcpounte Immune score among molecular subtypes in TCGA dataset; c Comparison of Estimate Immune score among molecular subtypes in TCGA dataset; d Comparison of different immune immune scores among molecular subtypes in TCGA dataset

Fig. 5

a Sankey map between our molecular subtype and existing subtypes; b Distribution comparison of CMS subtypes between different molecular subtype; c Distribution comparison of Thorsson subtypes between different molecular subtype

Fig. 6

a The changing trajectory of each independent variable. The horizontal axis represents the log value of the independent variable lambda, and the vertical axis represents the coefficient of the independent variable. b The confidence interval under each lambda

Fig. 7

a Survival curves between two risk groups based on 6-gene signature classification; b distribution of RiskScore and survival status of 6-gene signature in TCGA training cohort; c ROC curve of 6-gene signature classification in TCGA training cohort; d–f survival curves between two risk groups, distribution of RiskScore and survival status, ROC curve of 6-gene signature in TCGA testing cohort; g–i survival curves between two risk groups, distribution of RiskScore and survival status, ROC curve of 6-gene signature in entire TCGA cohort; j–l survival curves between two risk groups, distribution of RiskScore and survival status, ROC curve of 6-gene signature in GSE17537cohort

Fig. 8

The prognostic performance of the risk model in different clinical features

Fig. 9

a comparison of the risk score in T Stage grouping samples; b comparison of the risk score in N Stage grouping samples; c comparison of the risk score in M Stage grouping samples; d comparison of the risk score in clinical stage grouping samples; e comparison of the risk score in CMS Subtypes; f comparison of the risk score in our molecular subtypes; g correlation between RiskScore with KKEGG_APOPTOSI; h correlation between RiskScore with KEGG_NOD_LIKE_ RECEPTOR_SIGNALING_PATHWAY

Figure10

a Univariate analysis of clinical features and RiskScore; b multivariate analysis of clinical features and RiskScore; c construction of nomogram model; d the Calibration curves of 1-, 3-, 5- year in nomogram; e DCA analysis of Age, M Stage, clinical Stage, risk score and nomogram

Fig. 11

a, b The ROC of 15-gene signature (Xu) risk model and KM curves of the High/Low COAD samples; c, d the ROC of 15-gene signature (Dai) risk model and KM curves of the High/Low COAD samples; e, f the ROC of 12-gene signature (Sun) risk model and KM curves of the High/Low COAD samples; g, h the ROC of 9-gene signature (Mo) risk model and KM curves of the High/Low COAD samples

Fig. 12

Correlation between the 8 genes’ expression and immune cell infiltration score; a–g correlation between the expression of HOXD9, ITLN1, TSPAN1, GPRC5B, CXCL13, TIMP1 with B cell, CD8+ T cell, CD4+ T cell, Macrophage, and Dendritic Cell, respectively. The horizontal axis refers to the infiltration level, and the vertical axis represents gene expression

Fig. 13

Expression box diagram of gene expression in pan-cancer. a–f The gene expression of HOXD9, CPRC5B, ITLN1, CXCL13, TSPAN11 and TIMP1 in different tumors, respectively

Fig. 14

Clinical validation of 6 genes in protein and mRNA expression. a The expression box graph of 6 genes in TCGA-COAD; b the expression box graph of 6 genes in GSE10972; c ITLN1 protein expression in cancer and normal control; d CXCL13 protein expression in cancer and normal control; e TIMP1 protein expression in cancer and a normal control; f Westernblot expression of 6 genes in 3 pairs of cancer and adjacent normal tissues; g–l mRNA expression of GPRC5B, HOXD, TSPAN1, ITLN1, TIMP1 and CXCL13, respectively in 3 pairs of cancer and adjacent normal tissues