An assessment of prognostic immunity markers in breast cancer

Abstract

Tumor-infiltrating lymphocytes (TIL) and immunity gene signatures have been reported to be significantly prognostic in breast cancer but have not yet been applied for calculation of risk of recurrence in clinical assays. A compact set of 17 immunity genes was derived herein from an Affymetrix-derived gene expression dataset including 1951 patients (AFFY1951). The 17 immunity genes demonstrated significant prognostic stratification of estrogen receptor (ER)-negative breast cancer patients with high proliferation gene expression. Further analysis of blood and breast cancer single-cell RNA-seq datasets revealed that the 17 immunity genes were derived from TIL that were inactive in the blood and became active in tumor tissue. Expression of the 17 immunity genes was significantly (p < 2.2E-16, n = 91) correlated with TILs percentage on H&E in triple negative breast cancer. To demonstrate the impact of tumor immunity genes on prognosis, we built a Cox model to incorporate breast cancer subtypes, proliferation score and immunity score (72 gene panel) with significant prediction of outcomes (p < 0.0001, n = 1951). The 72 gene panel and its risk evaluation model were validated in two other published gene expression datasets including Illumina beads array data METABRIC (p < 0.0001, n = 1997) and whole transcriptomic mRNA-seq data TCGA (p = 0.00019, n = 996) and in our own targeted RNA-seq data TARGETSEQ (p < 0.0001, n = 303). Further examination of the 72 gene panel in single cell RNA-seq of tumors demonstrated tumor heterogeneity with more than two subtypes observed in each tumor. In conclusion, immunity gene expression was an important parameter for prognosis and should be incorporated into current multi-gene assays to improve assessment of risk of distant metastasis in breast cancer.

Conflict of interest statement

The Authors declare no competing interests.

Figures

Fig. 1

RNA-seq gene expression of 17 immunity genes and 19 proliferation genes in published PBMC single-cell dataset (Macosko et al. Cell 2015) and breast cancer solid tumor single cells (Chung et al. Nature Communications 2017). (A) Expression of immunity and proliferation genes in different PBMC cell types including B cells, CD4 T cells, CD8 T cells, NK cells, Monocytes and dendritic cells. (B) Expression of immunity and proliferation genes of single cells including breast tumors’ immune single cells labeled as T_cell + Mac + M & B_cell (Mac = macrophages, M = monocytes) and carcinoma single cells groups 1 to 3 (mixed carcinoma single cells from different tumors), BC02 and BC05 (carcinoma single cells from each individual tumor)

Fig. 2

Survival plots of Immunity Score in different patient groups identified by proliferation and ER status in the AFFY1951 training dataset, two test datasets METABRIC and TARGETSEQ. Immunity Score demonstrated strongest outcome prediction in patients who were ER-negative and proliferation high in AFFY1951 (A) (p < 0.0001, n = 467), METABRIC (C) (p = 0.00018, n = 398) and TARGETSEQ (E) (p = 0.044, n = 65), but was insignificant in ER-positive and proliferation high patients in AFFY1951 (B) (p = 0.079, n = 585), METABRIC (D) (p = 0.22, n = 603) and TARGETSEQ (F) (p = 0.46, n = 97). High proliferation groups had proliferation scores no less than 50 and low proliferation groups had proliferation scores less than 50. Survival analysis of ER-negative or ER-positive and low proliferation patients were demonstrated in Supplementary Figure 1

Fig. 3

Comparison of survival analysis of iRDM and PAM50 in AFFY1951 breast cancer training dataset. Subtypes and risk groups are color-coded: Basal-like (Red), HER2E (Hot Pink), Immuno (Yellow), Luminal A (Dark Blue), Luminal B (Sky Blue), Normal-like (Green); low (Green), med (Dark Blue), and high (Red) risks. Kaplan-Meier plots were used to show Distant Metastasis-Free Survival (DMFS) by subtypes for iRDM (A) (p < 0.0001, n = 1655) and PAM50 (B) (p < 0.0001, n = 1463) and risk groups for iRDM (C) (p < 0.0001, n = 1951) and PAM50 (D) (p < 0.0001, n = 1951)

Fig. 4

Validation of iRDM subtype and risk survival analysis in three independent test datasets. Survival plots of iRDM subtypes in METABRIC (A) (p < 0.0001, n = 1997), TCGA (C) (p = 0.022, n = 996), TARGETSEQ (E) (p = 0.00021, n = 303) and survival plots of risk groups (high, low, med) in METABRIC (B) (p < 0.0001, n = 1997), TCGA (D) (p = 0.00019, n = 996), TARGETSEQ (E) (p < 0.0001, n = 303) were shown. Subtypes and risk groups are color-coded: Basal-like (Red), HER2E (Hot Pink), Immuno (Yellow), Luminal A (Dark Blue), Luminal B (Sky Blue), Normal-like (Green); low (Green), med (Dark Blue), and high (Red) risks

Fig. 5

Heatmap of iRDM subtypes analyzed in 549 single cells from 11 primary breast tumors and two lymph node metastases. All tumors showed two or more iRDM subtypes. Sub, subtype; Cell, single cell number; TIL, percentage of tumor-infiltrating lymphocytes. Color-coded individual cell subtypes: Basal-like (Red), HER2E (Hot Pink), Immuno (Yellow), Luminal A (Dark Blue), Luminal B (Sky Blue), Normal-like (Green), Mixed (Black)