Molecular subtype construction and prognosis model for stomach adenocarcinoma characterized by metabolism-related genes

Abstract

Background: Metabolic reprogramming is implicated in cancer progression. However, the impact of metabolism-associated genes in stomach adenocarcinomas (STAD) has not been thoroughly reviewed. Herein, we characterized metabolic transcription-correlated STAD subtypes and evaluated a metabolic RiskScore for evaluation survival.

Method: Genes related to metabolism were gathered from previous study and metabolic subtypes were screened using ConsensusClusterPlus in TCGA-STAD and GSE66229 dataset. The ssGSEA, MCP-Count, ESTIMATE and CIBERSORT determined the immune infiltration. A RiskScore model was established using the WGCNA and LASSO Cox regression in the TCGA-STAD queue and verified in the GSE66229 datasets. RT-qPCR was employed to measure the mRNA expressions of genes in the model.

Result: Two metabolism-related subtypes (C1 and C2) of STAD were constructed on account of the expression profiles of 113 prognostic metabolism genes with different immune outcomes and apparently distinct metabolic characteristic. The overall survival (OS) of C2 subtype was shorter than that of C1 subtype. Four metabolism-associated genes in turquoise model, which closely associated with C2 subtype, were employed to build the RiskScore (MATN3, OSBPL1A, SERPINE1, CPNE8) in TCGA-train dataset. Patients developed a poorer prognosis if they had a high RiskScore than having a low RiskScore. The promising effect of RiskScore was verified in the TCGA-test, TCGA-STAD and GSE66229 datasets. The prediction reliability of the RiskScore was validated by time-dependent receiver operating characteristic curve (ROC) and nomogram. Moreover, samples with high RiskScore had an enhanced immune status and TIDE score. Moreover, MATN3, OSBPL1A, SERPINE1 and CPNE8 mRNA levels were all elevated in SGC7901 cells. Inhibition of OSBPL1A decreased SGC7901 cells invasion numbers.

Conclusion: This work provided a new perspective into heterogeneity in metabolism and its association with immune escape in STAD. RiskScore was considered to be a strong prognostic label that could help individualize the treatment of STAD patients.

Keywords: Genes; Immune landscape; Metabolism; Prognosis; Prognosis model; Stomach adenocarcinomas.

Fig. 1

2 clusters were identified based on metabolism related genes. (A) 47 metabolic-related differentially expressed genes were screened. (B) Heatmap of clustering the samples in TCGA-STAD dataset when k = 2. (C) Heatmap of sample clustering in GSE66229 dataset when k = 2. (D) KM survival curve of 2 clusters in TCGA-STAD dataset. (E) KM survival curve of 2 clusters in GSE66229 dataset.

Fig. 2

The expression differences of 47 metabolic-related differentially expressed genes and distribution differences of clinical features between 2 clusters.

Fig. 3

113 metabolic pathways scores differences between 2 clusters.

Fig. 4

Immunoinfiltration analysis between 2 clusters. (A) C2 subtype had higher StromalScore. (B) C2 subtype had enhanced ImmuneScore. (C) 22 kind immune cells score differences between 2 clusters calculated by CIBERSORT. (D) Heat map of MCP-count and ssGSEA for calculating immunity scores between 2 clusters. (E) TIDE score differences between 2 clusters. (F) GSEA analysis in 2 clusters in TCGA-STAD dataset. (ns, no significant; *p < 0.05; **p < 0.01; ***p < 0.001; and ****p < 0.0001).

Fig. 5

Analysis of somatic mutations. (A) Comparisons on the number of Segments, Fraction Altered, TMB, and Aneuploidy Score, Homologous Recombination Defects in 2 clusters were compared. (B) Top15 gene mutation in 2 clusters.

Fig. 6

Turquoise module was a hub module. (A) A cluster tree for TCGA-STAD sample. Various soft-thresholding powers calculated by analyzing the scale-free fit index (β); The mean connectivity to determine different soft-thresholding powers. (B) Based on 1-TOM, differentially expressed genes were clustered and visualized in dendrogram. (C) Correlation between 9 modules and two clusters. (D) Number of genes in 9 modules. (E) Scatter diagram was plotted based on module membership vs. gene significance for C2 in the turquoise module.

Fig. 7

Signature development using the 4 genes. (A) The RiskScore prediction in TCGA training dataset was assess by the KM and ROC curves. (B) The RiskScore prediction in TCGA test dataset was assess by the KM and ROC curves. (C) The RiskScore prediction in entire TCGA dataset was assess by the KM and ROC curves. (D) The RiskScore prediction in GSE66229 dataset was assess by the KM and ROC curves. (E) Multivariate forest map of genes in prognosis model. (F) Univariate cox analysis on the RiskScore and clinical parameters. (G) Multivariate cox analysis on the RiskScore and clinical parameters. (H) A nomogram developed by clinical parameters of independent prognosis and RiskScore. (I) 1-, 3-, 5- year calibration curve of nomogram. (J) The decision curve for the nomogram, RiskScore and independent prognostic clinical features.

Fig. 8

Genes in model was validated. (A–D) MATN3, OSBPL1A, SERPINE1 and CPNE8 mRNA levels were elevated in SGC7901 cells than normal GSE-1 cells. (E) Inhibition of OSBPL1A decreased SGC7901 cells invasion numbers. (**p < 0.01; ***p < 0.001; and ****p < 0.0001).

Fig. 9

Correlation analysis between RiskScore and clinical features. (A) The RiskScore differences among StageⅠ-Ⅳ. (B) The RiskScore differences among Grade 1-3. (C) The RiskScore differences between C1 and C2. (D) the expressions of 4 genes in RiskScore and clinical feature in high- and low-group. (ns, no significant; ****p < 0.0001).

Fig. 10

Correlation analysis between RiskScore and immune cells. (A) Differences in TIDE scores between the two groups. (B) Differences in StromalScore, ImmuneScore and EstmateScore between the two groups. (C) ESTIMATE, MCP-Count and ssGSEA analysis were used to perform correlation analysis between immune cells and RiskScore. (**p < 0.01; ****p < 0.0001).