Development of an Immune-Related Prognostic Signature in Breast Cancer

Abstract

Background: Although increased early detection, diagnosis and treatment have improved the outcome of breast cancer patients, prognosis estimation still poses challenges due to the disease heterogeneity. Accumulating data indicated an evident correlation between tumor immune microenvironment and clinical outcomes.

Objective: To construct an immune-related signature that can estimate disease prognosis and patient survival in breast cancer.

Methods: Gene expression profiles and clinical data of breast cancer patients were collected from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases, which were further divided into a training set (n = 499), a testing set (n = 234) and a Meta-validation set (n = 519). In the training set, immune-related genes were recognized using combination of gene expression data and ESTIMATE algorithm-derived immune scores. An immune-related prognostic signature was generated with LASSO Cox regression analysis. The prognostic value of the signature was validated in the testing set and the Meta-validation set.

Results: A total of 991 immune-related genes were identified. Twelve genes with non-zero coefficients in LASSO analysis were used to construct an immune-related prognostic signature. The 12-gene signature significantly stratified patients into high and low immune risk groups in terms of overall survival independent of clinical and pathologic factors. The signature also significantly stratified overall survival in clinical defined groups, including stage I/II disease. Several biological processes, such as immune response, were enriched among genes in the immune-related signature. The percentage of M₂ macrophage infiltration was significantly different between low and high immune risk groups. Time-dependent ROC curves indicated good performance of our signature in predicting the 1-, 3- and 5-year overall survival for patients from the full TCGA cohort. Furthermore, the composite signature derived by integrating immune-related signature with clinical factors, provided a more accurate estimation of survival relative to molecular signature alone.

Conclusion: We developed a 12-gene prognostic signature, providing novel insights into the identification of breast cancer with a high risk of death and assessment of the possibility of immunotherapy incorporation in personalized breast cancer management.

Keywords: breast cancer; immune system; model statistical; prognosis; survival.

Figure 1

Immune scores and stromal scores in breast cancer. Violin-plot showing the distribution of immune scores (A) and stromal scores (B) among different breast cancer molecular subtypes. Kaplan−Meier curves of overall survival (OS) among high or low risk groups based on immune scores (C) and stromal scores (D) in the TCGA cohort.

Figure 2

Comparison of gene expression profile with immune scores in breast cancer. (A) Volcano plot visualizing the immune-related genes between high and low immune score groups. Red plots represent aberrantly expressed genes with adjusted P-value < 0.01 and log₂FC > 1. Black plots represent normally expressed genes. Blue plots represent aberrantly expressed genes with adjusted P value < 0.01 and log₂FC < 1. (B) Heatmap analysis of the immune-related genes between high and low immune score groups. (C) Forest plot of hazard ratios showing the survival-associated immune-related genes.

Figure 3

12-gene prognostic signature biomarker characteristics in TCGA training set, TCGA testing set and Meta-validation set. (A), (D) and (G) Breast cancer were ranked by immune risk scores in the TCGA training set, TCGA testing set and Meta-validation set. (B), (E) and (H) Heatmap of 12 genes related to IRGPI differentially expressed between high and low immune risk groups in the TCGA training set, TCGA testing set and Meta-validation set, with red indicating higher expression and blue indicating lower expression. Patients were stratified by immune-related gene prognostic index (IRGPI) (low immune risk vs. high immune risk). (C), (F), and (I) Kaplan−Meier curves of overall survival (OS) among different IRGPI risk groups in TCGA training set, TCGA testing set and meta-validation set. Hazard ratios (HRs) and 95% CIs are for high immune risk group vs. low immune risk group. P values comparing risk groups were calculated with the log-rank test.

Figure 4

Kaplan−Meier curve of overall survival (OS) for breast cancer patients with different IRGPI risks. (A) Survival curve of stage I/II patients of TCGA cohort. (B) Survival curve of stage I/II patients in GSE20685 dataset. (C) Survival curve of stage I patients of TCGA cohort. (D) Survival curve of stage I patients of GSE20685 dataset. (E) Survival curve of stage II patients of TCGA cohort. (F) Survival curve of stage II patients of GSE20685 dataset.

Figure 5

Comparison of our 12-gene signature and other models. Time-dependent ROC analysis was performed to compare our 12-gene signature and Oncotype Dx in predicting 1-year (A), 3-year (B) and 5-year (C) overall survival (OS). Time-dependent ROC analysis was performed to compare our 12-gene signature and other two immune-related signatures in predicting 1-year (D), 3-year (E) and 5-year (F) overall survival (OS).

Figure 6

(A) The difference of IRGPI between pCR and non-pCR groups in GSE32646 and GSE28844. (B) The difference of IRGPI between pCR and non-pCR groups in GSE16446 and GSE4779. (C) 22 types of immune cells abundance calculated by CIBERSORT in TCGA cohort. (D) Immune cells abundance between low and high immune risk groups for TCGA cohort. (E) B cells naïve, Macrophages M₂, T cells follicular helper and Macrophages M₀ abundance in GSE20685 dataset. The association between adjusted IRGPI and the abundance of DCs (F), B cells (G), T cells CD4 (H), macrophages (I), neutrophils (J), and T cells CD8 (K) calculated by TIMER in TCGA cohort. P values are based on Wilcoxon rank sum test. Error bars indicate estimated 95%CI.

Figure 7

Immune-clinical prognostic signature characteristics in TCGA training set, TCGA testing set and GSE20685 validation set. (A), (D), and (G) Breast cancer were ranked by immune-clinical risk scores in the TCGA training set, TCGA testing set and GSE20685 validation set. (B), (E), and (H) Kaplan−Meier curves of overall survival (OS) among different ICPI risk groups in TCGA training set, TCGA testing set and GSE20685 validation set. Hazard ratios (HRs) and 95% CIs are for high vs. low immune risk. P values comparing risk groups were calculated with the log-rank test. (C), (F), and (I) Restricted mean survival curves for IRGPI and ICPI scores was plotted on TCGA training set, TCGA testing set and GSE20685 validation set.

Figure 8

The workflow of construction and validation of our immune-related prognostic signature.