Network Modularity in Breast Cancer Molecular Subtypes

Abstract

Breast cancer is a heterogeneous and complex disease, a clear manifestation of this is its classification into different molecular subtypes. On the other hand, gene transcriptional networks may exhibit different modular structures that can be related to known biological processes. Thus, modular structures in transcriptional networks may be seen as manifestations of regulatory structures that tightly controls biological processes. In this work, we identify modular structures on gene transcriptional networks previously inferred from microarray data of molecular subtypes of breast cancer: luminal A, luminal B, basal, and HER2-enriched. We analyzed the modules (communities) found in each network to identify particular biological functions (described in the Gene Ontology database) associated to them. We further explored these modules and their associated functions to identify common and unique features that could allow a better level of description of breast cancer, particularly in the basal-like subtype, the most aggressive and poor prognosis manifestation. Our findings related to the immune system and a decrease in cell death-related processes in basal subtype could help to understand it and design strategies for its treatment.

Keywords: Functional modules; breast cancer subtypes; community structure; gene regulatory networks (GRN); network modularity; pathway enrichment analysis.

Figure 1

Workflow followed in this study. The pipeline starts with the collection of 493 breast cancer samples. PAM50 classification was performed as in (de Anda-Jáuregui et al., 2015). In the image, the colors correspond to each molecular subtype. For the four subtypes, mutual information-based networks (de Anda-Jáuregui et al., 2016) were inferred with ARACNE (Margolin et al., 2006). Once networks were built, we detected communities for each subtype network by using Infomap (Rosvall and Bergstrom, 2008). Modules are represented by different colored nodes. Detected modules were later enriched using HTSanalyzeR (Wang et al., 2011). In the picture, enriched communities are full colored circles, meanwhile non-enriched ones are non colored. Finally, the enriched modules were analyzed at a more profound level, by observing the most general processes that are involved in these modules. In the picture, enriched communities found in basal subtype are classified according to upper general processes. Each process is represented with a different color and the labels correspond to the GO-ID.

Figure 2

Modules in networks for each breast cancer molecular subtype. (A) Luminal A; (B) Luminal B; (C) HER2+ and (D) Basal subtype. The nodes belonging to a community are colored the same. The same color for different subtypes is not related. For visualization purposes, only islands with more than 8 genes are depicted.

Figure 3

COL5A2 communities: Enriched processes are shared between subtypes despite gene compositions being different. (A) Venn diagram showing the number of genes of COL5A2_comm for each molecular subtype. Notice that only 3 genes (COL5A2, THBS2, and LUM) are shared. (B) Enriched processes of COL5A2_comm for each molecular subtype.

Figure 4

Genes in COL5A2 modules are mostly overexpressed thruoghout molecular subtypes This figure shows the expression signature of those genes belonging to COL5A2 modules in (A) Luminal A (B), Luminal B (C), HER2+, and (D) Basal breast cancer molecular subtypes. Notice that the majority of genes are overexpressed. However, in (C) there is a subset of genes which is underexpressed and are grouped in terms of the network topology.

Figure 5

Modular structure of Basal subtype islands 2 and 6. (A) Colors define each module. (B) Information flow between communities. Link width is proportional to the number of links shared between modules. Full-color communities represent those that are enriched to a GO category.

Figure 6

GO processes associated with modules in the transcriptional network of basal breast cancer molecular subtype. In this figure, modules are colored according to the color code of Figure 5. These communities are connected to GO ID categories which are colored according to a general process (upper left). Names of those categories are provided in Supplementary Material 5.

Figure 7

Cell death and viral infection processes are oppositely regulated by the same molecular signature of basal breast cancer subtype. In this figure, genes are depicted according to their expression levels: red for overexpressed and blue for underexpressed genes. Lines between molecules and processes indicate the predicted function of the molecule according to its expression value, blue line leads an inhibition of the process; in turn, orange lines account for predicted activation. Color of cell death and viral infection processes represent the same predicted effect than lines.