Gene-expression molecular subtyping of triple-negative breast cancer tumours: importance of immune response

Abstract

Introduction: Triple-negative breast cancers need to be refined in order to identify therapeutic subgroups of patients.

Methods: We conducted an unsupervised analysis of microarray gene-expression profiles of 107 triple-negative breast cancer patients and undertook robust functional annotation of the molecular entities found by means of numerous approaches including immunohistochemistry and gene-expression signatures. A triple-negative external cohort (n=87) was used for validation.

Results: Fuzzy clustering separated triple-negative tumours into three clusters: C1 (22.4%), C2 (44.9%) and C3 (32.7%). C1 patients were older (mean=64.6 years) than C2 (mean=56.8 years; P=0.03) and C3 patients (mean=51.9 years; P=0.0004). Histological grade and Nottingham prognostic index were higher in C2 and C3 than in C1 (P<0.0001 for both comparisons). Significant event-free survival (P=0.03) was found according to cluster membership: patients belonging to C3 had a better outcome than patients in C1 (P=0.01) and C2 (P=0.02). Event-free survival analysis results were confirmed when our cohort was pooled with the external cohort (n=194; P=0.01). Functional annotation showed that 22% of triple-negative patients were not basal-like (C1). C1 was enriched in luminal subtypes and positive androgen receptor (luminal androgen receptor). C2 could be considered as an almost pure basal-like cluster. C3, enriched in basal-like subtypes but to a lesser extent, included 26% of claudin-low subtypes. Dissection of immune response showed that high immune response and low M2-like macrophages were a hallmark of C3, and that these patients had a better event-free survival than C2 patients, characterized by low immune response and high M2-like macrophages: P=0.02 for our cohort, and P=0.03 for pooled cohorts.

Conclusions: We identified three subtypes of triple-negative patients: luminal androgen receptor (22%), basal-like with low immune response and high M2-like macrophages (45%), and basal-enriched with high immune response and low M2-like macrophages (33%). We noted out that macrophages and other immune effectors offer a variety of therapeutic targets in breast cancer, and particularly in triple-negative basal-like tumours. Furthermore, we showed that CK5 antibody was better suited than CK5/6 antibody to subtype triple-negative patients.

Figure 1

Fuzzy clustering of 107 triple-negative breast cancer patients. Distribution of patients based on probability of belonging to cluster C1 (blue), C2 (red) and C3 (green). Each vertex of the triangle represents a cluster and each point represents a patient, placed as the barycentre of the triangle, weights being the probabilities of belonging to each of the clusters. Hence, the closer a point is to one of the vertices, the greater is the probability of the patient belonging to the corresponding cluster.

Figure 2

Kaplan-Meier analyses of 107 triple-negative breast cancer patients based on fuzzy-clustering partition. (A) Overall survival (OS) analysis shows that C3 patients have a better outcome than C1 (P = 0.0217) and C2 patients (P = 0.05). (B) Event-free survival (EFS) analysis shows the same result: C3/C1, P = 0.0145; C3/C2, P = 0.0195.

Figure 3

Functional annotations of the clustering results of our cohort. (A) Fuzzy-clustering probability of belonging to clusters, from 0 to 1. (B) Cluster numbering scheme. (C) Clinicopathologic characteristics with significant differences between clusters. Age as a continuous colour scale, from 28 years old (pale turquoise) to 85 years old (dark blue); Elston-Ellis (EE) histological grades 1 (pale turquoise), 2 (deep sky blue), and 3 (dark blue); Nottingham prognostic index (NPI) 1 (pale turquoise), 2 (deep sky blue), and 3 (darkblue). (D) Immunohistochemistry results for ten markers: positive (brown) and negative (blue). (E) Molecular subtyping by means of nine gene expression signatures (GES): three single sample predictors (SSPs) - luminal A (dark blue), luminal B (sky blue), HER2-E (purple), basal-like (red), normal breast-like (green) or unclassified (yellow); proliferation (continuous colour scale from minimum (6.59; deep sky blue) to maximum (11.05; red)); TNBCtype (basal-like 1 (red), basal-like 2 (dark red), immunomodulatory (dark green), mesenchymal-like (pink), mesenchymal stem-like (magenta), luminal androgen receptor (blue) and unclassified tumours (yellow)); Teschendorff’s GES (cell cycle (blue), cell cycle and immune response (orange), extracellular matrix (green), immune response (red), steroid hormone response (pink) and unclassified (yellow)); claudin-low (claudin-low (yellow), other (darkblue)); vascular endothelial growth factor (VEGF; continuous colour scale, from minimum (8.13; deep sky blue) to maximum (10.67; red)); and glycolysis (continuous colour scale, from minimum (9.63; deepskyblue) to maximum (11.88; red)). (F) Gene ontology biological process terms enrichment. Missing values are in white.

Figure 4

C2/C3 immune response dissection. (A) Ward’s hierarchical clustering and heatmap showing the segregation of C2 (red) and C3 (green) patients as a function of the seven Rody’s metagenes (B lymphocytes (IgG); macrophages and monocyte/myeloid lineage cells (HCK); professional antigen-presenting cells (MHC-II); T-cell (LCK); cell types for presentation of intracellular antigens (MHC-I); interferon signal transduction (STAT1); interferon response (interferon)) and M2/M1 gene-expression signature (GES) (continuous colour scale, from minimum (8.05, 6.90, 5.06, 7.84, 7.43, 9.68, 11.19 and −1.36 for HCK, LCK, IgG, STAT1, Interferon, MHC-II, MHC-I and M2/M1 GES, respectively; deep sky blue) to maximum (11.46, 12.24, 12.93, 13.21, 12.55, 13.75, 14.69 and −0.68 for HCK, LCK, IgG, STAT1, Interferon, MHC-II, MHC-I and M2/M1 GES, respectively; red). (B) Kaplan-Meier curves for event-free survival (EFS) analysis of breast cancer patients with high immune response (HIR) and low immune response (LIR).

Figure 5

Functional annotations of the clustering results for triple-negative patients from GSE21653 ( n  = 87). (A) Fuzzy-clustering probability of belonging to clusters, from 0 to 1. (B) Cluster numbering scheme. (C) Clinicopathologic characteristics with significant differences between clusters: age as a continuous colour scale, from 28 years old (pale turquoise) to 85 years old (dark blue); Elston-Ellis (EE) histological grades 1 (pale turquoise), 2 (deep sky blue), and 3 (dark blue); Nottingham prognostic index (NPI) 1 (pale turquoise), 2 (deep sky blue), and 3 (dark blue). (D) Immunohistochemistry results for Ki67: positive (brown) and negative (blue). (E) Molecular subtyping by means of nine gene expression signatures (GES): three single sample predictors (SSPs) - luminal A (dark blue), luminal B (sky blue), HER2-E (purple), basal-like (red), normal breast-like (green) or unclassified (yellow); proliferation (continuous colour scale from minimum (6.01; deep sky blue) to maximum (10.92; red)); TNBCtype (basal-like 1 (red), basal-like 2 (dark red), immunomodulatory (dark green), mesenchymal-like (pink), mesenchymal stem-like (magenta), luminal androgen receptor (blue) and unclassified tumours (yellow)); Teschendorff’s GES (cell cycle (blue), cell cycle and immune response (orange), extracellular matrix (green), immune response (red), steroid hormone response (pink) and unclassified (yellow)); claudin-low (claudin-low (yellow), other (dark blue)); vascular endothelial growth factor (VEGF; continuous colour scale, from minimum (8.12; deep sky blue) to maximum (10.82; red)); and glycolysis (continuous colour scale, from minimum (9.42; deep sky blue) to maximum (13.00; red)). (F) Gene ontology biological process terms enrichment. Missing values are in white.