The New Paradigm of Network Medicine to Analyze Breast Cancer Phenotypes

Abstract

Breast cancer (BC) is a heterogeneous and complex disease as witnessed by the existence of different subtypes and clinical characteristics that poses significant challenges in disease management. The complexity of this tumor may rely on the highly interconnected nature of the various biological processes as stated by the new paradigm of Network Medicine. We explored The Cancer Genome Atlas (TCGA)-BRCA data set, by applying the network-based algorithm named SWItch Miner, and mapping the findings on the human interactome to capture the molecular interconnections associated with the disease modules. To characterize BC phenotypes, we constructed protein-protein interaction modules based on "hub genes", called switch genes, both common and specific to the four tumor subtypes. Transcriptomic profiles of patients were stratified according to both clinical (immunohistochemistry) and genetic (PAM50) classifications. 266 and 372 switch genes were identified from immunohistochemistry and PAM50 classifications, respectively. Moreover, the identified switch genes were functionally characterized to select an interconnected pathway of disease genes. By intersecting the common switch genes of the two classifications, we selected a unique signature of 28 disease genes that were BC subtype-independent and classification subtype-independent. Data were validated both in vitro (10 BC cell lines) and ex vivo (66 BC tissues) experiments. Results showed that four of these hub proteins (AURKA, CDC45, ESPL1, and RAD54L) were over-expressed in all tumor subtypes. Moreover, the inhibition of one of the identified switch genes (AURKA) similarly affected all BC subtypes. In conclusion, using a network-based approach, we identified a common BC disease module which might reflect its pathological signature, suggesting a new vision to face with the disease heterogeneity.

Keywords: Disease modules; Network Medicine; Network-based algorithm; Switch genes and Interactome; TCGA; breast cancer.

Figure 1

Bioinformatic workflow. Study design for the identification of breast cancer switch genes from IHC and PAM50 subtype classification. (A) Computational analysis; (B) Integrated functional analysis; (C) Integration of IHC and PAM50 Key Regulatory Switches. pos: positive; neg: negative.

Figure 2

Heat cartography maps for each breast subtype. The cartographic representation of the correlation network of each analyzed breast subtype is reported. The x and y axes correspond to the within-module degree zg and the clusterphobic coefficient Kp, respectively. The within-module degree zg measures how “well-connected” each node is to other nodes in its own cluster, and the clusterphobic coefficient Kp measures the “fear” of being confined in a cluster. High zg values correspond to nodes that are hubs within their community, while high values of Kp identify nodes that interact mainly outside their community, i.e., with many more external than internal links [8]. Dots correspond to nodes in the networks colored according to the value of the Average Pearson Correlation Coefficients (APCC) between its expression profile and that of its nearest neighbors. Switch genes correspond to nodes colored in blue and falling in R4 region (highlighted with a red circle).

Figure 3

Overlapping BC subtype PPI-based modules by switch genes. (A) Schematic representation of the overlapping modules identified by BC subtypes switch genes in the human interactome. (B,C) Bar plot reporting the values of the network-based separation measure computed between each pair of switch genes’ modules of PAM50 classification (B) and IHC classification (C).

Figure 4

Integration and prediction of IHC and PAM50 key regulatory switches. (A) and (B) Heatmaps of the enriched pathways for IHC and PAM50 classification, respectively. Enrichment score (Fisher’s Exact right-tailed test, p-value < 0.01). (C) Prediction of inhibited (blue squares) and activated (orange squares) functions of shared IHC and PAM50 switches sorted by activation Z-score (absolute value 3). (D) Venn diagram defining the Intersected Shared (IS) switches of shared IHC and PAM50 switches. (E) Protein–protein interaction (PPI) network of the 28 Unique Shared IHC–PAM50 switches resulted in 24 PPI network molecules. (F) MCODE1 represents a densely connected protein complex (6 in red out of 24) of the PPI network by the Molecular Complex Detection algorithm.

Figure 5

Relative expression of switch genes in BC cell lines and tissue. (A) Switch gene expression was evaluated in BC cell lines vs control cell by real-time PCR. (B) AURKA, CDC45, ESPL1, and RAD54L genes were evaluated in BC tissue using TissueScan qPCR Arrays. *p < 0.05 was considered statistically significant.

Figure 6

AURKA expression in vitro and ex vivo. (A) AURKA was examined by immunoblotting in T47D, MCF7 (Lum A), BT474, MDA-MB-361 (Lum B), MDA-MB-453, SK-BR3, UACC-893 (Her2), BT-549, Hs-578T (Triple Negative) and MCF10a (Control). Β-actin served as a loading control. (B) Immunohistochemical analysis of AURKA was performed in surgical specimens of BC. The panel shows a control tissue with no AURKA-positive cells, and four tumoral tissues, one for each subtype (Lum A, Lum B, Her2, Triple-negative) with AURKA-positive cells. Magnification 20X.

Figure 7

Effect of Alisertib treatment on BC cell lines. (A) Alisertib effects were assessed on cell growth: T-47D, BT-474, SKBR3, and MDA-MB-231 cell lines were treated with 0.25, 0.5, 1 and 2 µM of the drug for 24, 48, and 72 h and DMSO treatment was used as a negative control. (B) Alisertib effects were assessed on cell cycle: T-47D, BT-474, SKBR3, and MDA-MB-231 cell lines were treated with 0.5 µM of the drug for 24, 48, and 72 h and DMSO treatment was used as a negative control.