Bulk and single-cell transcriptome profiling reveal the metabolic heterogeneity in gastric cancer

Abstract

Metabolic reprogramming has been defined as a key hall mark of human tumors. However, metabolic heterogeneity in gastric cancer has not been elucidated. Here we separated the TCGA-STAD dataset into two metabolic subtypes. The differences between subtypes were elaborated in terms of transcriptomics, genomics, tumor-infiltrating cells, and single-cell resolution. We found that metabolic subtype 1 is predominantly characterized by low metabolism, high immune cell infiltration. Subtype 2 is mainly characterized by high metabolism and low immune cell infiltration. From single-cell resolution, we found that the high metabolism of subtype 2 is dominated by epithelial cells. Not only epithelial cells, but also various immune cells and stromal cells showed high metabolism in subtype 2 and low metabolism in subtype 1. Our study established a classification of gastric cancer metabolic subtypes and explored the differences between subtypes from multiple dimensions, especially the single-cell resolution.

Figure 1

Gastric cancers exhibit metabolic heterogeneity. (A–D) The consensus matrix shows patients with two metabolic subtypes in TCGA-STAD, GPL570 meta-dataset, GSE26942 and GSE84437. (E–H) Kaplan–Meier curves for overall survival based on metabolic subtypes (Log-rank test) in TCGA-STAD, GPL570 meta-dataset, GSE26942 and GSE84437. (I) KEGG enrichment results of 98 prognosis-related metabolic genes.

Figure 2

Analysis of signaling pathways between metabolic subtypes. (A) Five tumor metabolism-related pathway differences between two metabolic subtypes. (B) Ten oncogenic-related pathway differences between two metabolic subtypes. (C) Metabolic pathways in two metabolic subtypes using Gene set enrichment analysis. SIZE represents the number of genes in the corresponding gene set. NES represents corrected normalized enrichment score. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Figure 3

Differences between metabolic subtypes. (A) Top 20 mutated genes in all gastric cancer patients of TCGA-STAD cohort. (B) Top 20 differentially mutated genes in all gastric cancer patients of TCGA-STAD cohort. (C) Comparison of TCGA gastric cancer subtypes among two metabolic subtypes. (D) comparison of tumor mutation burden among two metabolic subtypes. (E) KEGG signaling pathway enriched for genes with low methylation and high expression in cluster 2. (F) KEGG signaling pathway enriched for genes with low methylation and high expression in cluster 1. (G) Stromal score, Immune score and ESTIMATE score between two metabolic subtypes. (H) Relative proportion of 22 infiltrating immune cells estimated by CIBERSORT between two metabolic subtypes of TCGA-STAD cohort. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Figure 4

The estimation of immunotherapy response, chemotherapy response and potential therapeutic drugs for gastric cancer. (A) TIDE scores of two metabolic subtypes in the TCGA-STAD cohort. (B) T cell dysfunction scores of two metabolic subtypes in the TCGA-STAD cohort. (C) T cell exclusion scores of two metabolic subtypes in the TCGA-STAD cohort. (D–H) The chemotherapy response of two metabolic subtypes for 5 common chemotherapy drugs.

Figure 5

Construction of metabolic subtype specific prognostic model. (A,B) Patients were divided into high-risk and low-risk subgroup based best cutoff, Kaplan–Meier analysis demonstrated that patients with higher metabolic subtype-associated signature score exhibited worse overall survival in TCGA-STAD, ROC curves showing the predictive efficiency of the model on the 1-, 3-, and 5-years survival rate. (C–E) the prognostic difference was validated in 3 independent cohorts.

Figure 6

Analysis of metabolic subtypes in the single-cell dimension. (A,B) Performance of metabolic subtype classification models on training and test datasets. (C) Quantitative distribution of the five major cell types between the two metabolic subtypes. (D) The UMAP plot of all cells, which are color-coded based on their associated clusters. (E) Differences among the three metabolic pathways enriched in cluster 1 at the bulk dimension among the five types of cells. (F–H) Differences among the 30 metabolic pathways enriched in cluster 2 at the bulk dimension among the five types of cells. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Figure 7

Metabolic pathway analysis of non-epithelial cells in single-cell sequencing. (A) Metabolic pathways analysis in T cells of different metabolic subtypes using Gene set enrichment analysis (Only metabolic subtype 2 had significantly enriched metabolic pathways). (B) Metabolic pathways analysis in B cells of different metabolic subtypes using Gene set enrichment analysis. (C) Metabolic pathways analysis in stromal cells of different metabolic subtypes using Gene set enrichment analysis. (D) Metabolic pathways analysis in myeloid cells of different metabolic subtypes using Gene set enrichment analysis (only metabolic subtype 2 had significantly enriched metabolic pathways).