Application of single-cell RNA sequencing analysis of novel breast cancer phenotypes based on the activation of ferroptosis-related genes

Abstract

Ferroptosis is distinct from classic apoptotic cell death characterized by the accumulation of reactive oxygen species (ROS) and lipid peroxides on the cell membrane. Increasing findings have demonstrated that ferroptosis plays an important role in cancer development, but the exploration of ferroptosis in breast cancer is limited. In our study, we aimed to establish a ferroptosis activation-related model based on the differentially expressed genes between a group exhibiting high ferroptosis activation and a group exhibiting low ferroptosis activation. By using machine learning to establish the model, we verified the accuracy and efficiency of our model in The Cancer Genome Atlas Breast Invasive Carcinoma (TCGA-BRCA) set and gene expression omnibus (GEO) dataset. Additionally, our research innovatively utilized single-cell RNA sequencing data to systematically reveal the microenvironment in the high and low FeAS groups, which demonstrated differences between the two groups from comprehensive aspects, including the activation condition of transcription factors, cell pseudotime features, cell communication, immune infiltration, chemotherapy efficiency, and potential drug resistance. In conclusion, different ferroptosis activation levels play a vital role in influencing the outcome of breast cancer patients and altering the tumor microenvironment in different molecular aspects. By analyzing differences in ferroptosis activation levels, our risk model is characterized by a good prognostic capacity in assessing the outcome of breast cancer patients, and the risk score can be used to prompt clinical treatment to prevent potential drug resistance. By identifying the different tumor microenvironment landscapes between the high- and low-risk groups, our risk model provides molecular insight into ferroptosis in breast cancer patients.

Keywords: Breast cancer; Ferroptosis; Single-cell RNA sequencing; Tumor microenvironment.

Fig. 1

Consensus clustering analysis of the breast cancer samples and prognosis between clusters. a Cluster breast cancer patients in the TCGA-BRCA set according to the expression level of ferroptosis-related genes. b The CDF curve was used to determine the best method of clustering. c Data used to measure the quality of clustering results. d The prognosis among the three clusters

Fig. 2

The expression of ferroptosis-related genes in the clusters and the prognostic features and clinical characteristics. a Heatmap of ferroptosis-related gene expression levels. b The result of PCA. c The prognostic difference between clusters

Fig. 3

Analysis of differentially expressed genes. a Volcano plot of differentially expressed genes. b Heatmap of the levels of differentially expressed genes

Fig. 4

The results of the univariate Cox analysis of prognosis-related genes. a Forest plot of the univariate Cox analysis result. b Genes with significant prognostic differences between the high ferroptosis-activated group and the low ferroptosis-activated group. c Expression level of the 14 prognosis-related genes using bioinformatics analysis

Fig. 5

The prognostic difference between FeAS groups. a The relationship of the expression of selected genes and clinical features in a heatmap. b The differences in OS between FeAS groups in the TCGA-BRCA set. c The ROC curve of 1-year, 2-year, and 3-year survival in the TCGA-BRCA set. d The differences in OS between FeAS groups in GSE96058. e The ROC curve of 1-year, 2-year, and 3-year survival in GSE96058

Fig. 6

Univariate and multivariate COX analysis. a Univariate COX analysis in the TCGA-BRCA set. b Multivariate COX analysis in the TCGA-BRCA set. c Univariate COX analysis in GSE96058. d Multivariate COX analysis in GSE96058

Fig. 7

Single-cell clustering analysis and pseudotime analysis in FeAS groups. a Cluster subtypes of breast cancer cells and subtypes based on CNV. b The distribution of FeAS groups in breast cancer cells. c The frequency distribution of the FeAS score in breast cancer cells. d The pseudotime trajectory in breast cancer cells

Fig. 8

The differentially expressed transcription factors between the high and low FeAS groups. a According to the synergy of different transcription factors, breast cancer samples were divided into 5 modules. b The potential relationship between the FeAS scoring system and the module. c Distribution of AUCs of different modules in breast cancer cells. d and e The first 10 differentially activated TFs in the high and low FeAS groups

Fig. 9

The enrichment distribution of the top 10 differentially activated TFs

Fig. 10

GO and KEGG enrichment analysis of the targeted genes of the top 10 differentially activated TFs

Fig. 11

The enrichment of immune-related cells or pathways and GSEA using bulk and single-cell RNA sequencing data between the high and low FeAS groups. a Enrichment analysis of immune-related cells or pathways in the bulk data. b Enrichment analysis of immune-related cells or pathways in single-cell RNA sequencing data. c GSEA of the bulk data. d GSEA of single-cell RNA sequencing data

Fig. 12

Immune infiltration landscape between the high and low FeAS groups. a–d The correlation between FeAS and stromal score, immune score, ESTIMATE score, and purity. e Analysis of immune cell infiltration based on ssGSEA

Fig. 13

Cell communication between cells in the high and low FeAS groups and different immune cells or signaling pathways. a–e The complex cell communication among high or low FeAS breast cancer cells and different immune cells. b The correlation between FeAS groups or immune cells and different signaling pathways. c SEMA4 signaling pathway network. d Incoming communication patterns of target cells. f Receptor ligands among different cells

Fig. 14

The treatment analysis between high and low FeAS groups based on IMvigor210. a Kaplan‒Meier analysis between FeAS groups. b The chemotherapy response between FeAS groups. c CR/PR and SD/PD between FeAS groups

Fig. 15

The correlation between FeAS and different drugs based on the GDSC and CCLE databases. a Drugs characterized by IC₅₀ positively correlated with FeAS score in the GDSC database. b Drugs characterized by IC₅₀ positively correlated with FeAS score in the CCLE database. c Drugs characterized by IC₅₀ negatively correlated with FeAS score in the CCLE database