Pan-cancer mapping of differential protein-protein interactions

Abstract

Deciphering the variations in the protein interactome is required to reach a systems-level understanding of tumorigenesis. To accomplish this task, we have considered the clinical and transcriptome data on >6000 samples from The Cancer Genome Atlas for 12 different cancers. Utilizing the gene expression levels as a proxy, we have identified the differential protein-protein interactions in each cancer type and presented a differential view of human protein interactome among the cancers. We clearly demonstrate that a certain fraction of proteins differentially interacts in the cancers, but there was no general protein interactome profile that applied to all cancers. The analysis also provided the characterization of differentially interacting proteins (DIPs) representing significant changes in their interaction patterns during tumorigenesis. In addition, DIP-centered protein modules with high diagnostic and prognostic performances were generated, which might potentially be valuable in not only understanding tumorigenesis, but also developing effective diagnosis, prognosis, and treatment strategies.

Figure 1

Schematic overview of differential interactome analysis between two phenotypes. (A) Highly probable protein-protein interactions (PPIs) in each phenotype. (B) Comparative analysis of interactome profiles of the phenotypes leads to the identification of differential PPIs, which are categorized into two groups as repressed or activated under the phenotype of interest (e.g., diseased phenotype). (C) Differentially interacting proteins (DIPs) representing significant changes in their interaction patterns during the transition between the phenotypes.

Figure 2

Differential interactome networks in 12 human cancers. Differential interactome network was constructed around dPPIs for each cancer. Red edges represent repressed interactions; black edges represent activated interactions.

Figure 3

Differential interactome in human cancers. (a) Bar chart showing the numbers of interactions being specific and non-specific to the cancer types and their count ratio in the 12 different cancer types. (b) Graph indicates the prevalence of dPPIs in different cancers. (c) Topological characteristics of differential interactome networks. Bean plots represent the distribution of topological metrics (i.e., clustering coefficient, network diameter, network centralization, characteristic path length, average number of neighbors, network heterogeneity, number of nodes, and number of interactions) across constructed differential interactome networks. The individual observations are shown as small lines.

Figure 4

Pan-cancer analysis of the differential interactome. Similarity network showing pairwise correlations among 12 cancer types based on the Jaccard indices.

Figure 5

Differential interacting protein networks. Network of DIPs in 12 cancers indicates the first five DIPs having the most interactions for each cancer in larger nodes. The DIPs observed in more than one cancer type are represented in white; the cancer-specific DIPs are represented in different colors.

Figure 6

Prognostic and principal component analyses for different cancer types. (a) Kaplan-Meier Plots estimating patients’ survival for 12 cancer types indicating p-value and hazard ratio for each curve. (b) PCA plots showing the individual differences in the gene expression profiles among the cancers including at least 30 individuals in each type.

Figure 7

Over-representation analysis for different cancer types. The graph shows the percentages of DIPs in various molecular functions to all genes participating in these molecular functions for different cancer types. Each molecular function is represented in different color and name of the cancers are indicated on the left going from the inner to the outer layer.

Figure 8

Classification of DIPs. The stacked bar graph indicates the number of DIPs, which are druggable (blue bars) and undruggable (orange bars) and their percentage for each cancer.