Dual proteome-scale networks reveal cell-specific remodeling of the human interactome

Abstract

Thousands of interactions assemble proteins into modules that impart spatial and functional organization to the cellular proteome. Through affinity-purification mass spectrometry, we have created two proteome-scale, cell-line-specific interaction networks. The first, BioPlex 3.0, results from affinity purification of 10,128 human proteins-half the proteome-in 293T cells and includes 118,162 interactions among 14,586 proteins. The second results from 5,522 immunoprecipitations in HCT116 cells. These networks model the interactome whose structure encodes protein function, localization, and complex membership. Comparison across cell lines validates thousands of interactions and reveals extensive customization. Whereas shared interactions reside in core complexes and involve essential proteins, cell-specific interactions link these complexes, "rewiring" subnetworks within each cell's interactome. Interactions covary among proteins of shared function as the proteome remodels to produce each cell's phenotype. Viewable interactively online through BioPlexExplorer, these networks define principles of proteome organization and enable unknown protein characterization.

Keywords: AP-MS; BioPlex; bioinformatics; cell specificity; computational biology; human interactome; network biology; protein interactions; proteomics; proteotypes.

Figure 1:. Interactome Profiling in Multiple Human Cell Lines

(A) Our ongoing effort to map the human interactome has culminated in BioPlex 3.0, which builds upon two prior versions and incorporates AP-MS experiments targeting 10,128 bait proteins in 293T cells. Bars beneath each network reflect the fraction of proteins targeted as baits. (B) We have repeated AP-MS analysis of 5,522 baits in HCT116 cells to produce a second proteome-scale interaction network. Bars beneath each network reflect the fraction of proteins targeted as baits. (C) Our latest networks in 293T and HCT116 cells expand coverage beyond previous attempts. Y2H: yeast-two-hybrid. Circle size is proportional to interaction count. (D) Comparison with BioGRID reveals that most BioPlex 3.0 interactions have not been previously reported. Incorporating PubMed citation counts for individual proteins suggests that much of the increased coverage comes from interactions among poorly studied proteins. (E) Overlap among proteins in 293T and HCT116 networks. (F) Overlap among interactions in 293T and HCT116 networks. A dashed box depicts the subset of 293T interactions matching those baits also targeted in HCT116 cells. (G) Ternary diagram depicting the proportions of edges shared or unique to either 293T or HCT116 cells for subnetworks defined by 912 CORUM complexes. Interactions observed among proteins in each complex were extracted from the combined 293T/HCT116 network and tallied to determine numbers of edges shared or specific to each cell line. Four individual complexes are displayed as Venn diagrams; each is also represented as a point within the ternary diagram whose location reflects the relative proportions of shared and cell-specific edges. Points near the corners indicate that most edges are either shared (“Both”) or cell-specific; points near the center of the triangle indicate edges evenly distributed across shared and cell-specific categories. The ternary diagram thus summarizes Venn diagrams for 912 complexes. A box-whisker plot depicts the edge overlap across complexes. (H) – (K) Subnetworks corresponding to four CORUM complexes highlighted in panel G.

Figure 2:. Structural Context Drives Interaction Replication in Protein Complexes

(A) Method for overlaying BioPlex interactions onto 3D structures and assessing detection and cellline specificity. See text for details. (B) The fraction of observable interactions detected in BioPlex networks and the fraction of BioPlex interactions that match direct interactions. Results aggregated across 309 PDB structures. (C) Relative odds that interactions are shared in 293T and HCT116 cells, versus cell-line-specific, as a function of inter-protein distance. (D) Ternary diagram depicts sharing of direct interactions in 293T and HCT116 networks across 306 PDB structures. (E) Ternary diagram depicts sharing of indirect interactions in 293T and HCT116 networks across 132 PDB structures with at least one indirect interaction. (F) – (H) Selected complexes. Each structure is displayed (Column 1) along with a network visualization of all direct interactions in the structure (Column 2). BioPlex edges are then overlayed to show which direct interactions were detected (Column 3) and to show direct and indirect edges colored according to whether they were detected in 293T, HCT116, or both (Column 4).

Figure 3:. Interactions within Complexes and Pathways Covary according to Shared Function and Cellular Phenotype

(A) Ternary diagrams depict edge sharing across cell lines for subnetworks defined by protein functional classes. Plots are shown for CORUM complexes, Reactome pathways, GO ontologies, and DisGeNET disease associations, ordered by edge sharing among constituent protein classes. Venn diagrams match assemblies shown in panels B, C, D, F, and H. (B) Glycolysis subnetwork (Reactome). (C) RNA Polymerase II Transcription Initiation subnetwork (Reactome). (D) EPH-Ephrin Signaling subnetwork (Reactome). (E) Phosphopeptides detected for ephrin receptors and ligands expressed as baits in each cell line. (F) “TGF-β Receptor Signaling Activates SMAD’s” subnetwork (Reactome). (G) Expression of epithelial and mesenchymal markers regulated by Snail, ZEB, and bHLH transcription factor family members downstream of TGF-β signaling. (H) Post-synaptic Membrane subnetwork (GO Cellular Component).

Figure 4:. Linking Differential TP53 Signaling to Cell-specific Interactions

(A) Cell cycle regulation: TP53 and the DREAM Complex. (B) Selected interactions of TP53 in 293T and HCT116 cells. (C) Enrichment of SV40 Large T Antigen in TP53 IP’s. (D) Expression of TP53 and related proteins in 293T and HCT116 cells. (E) HCT116-specific interactions involving proteins whose expression is regulated by TP53. (F) RBL1/2 abundance and phosphorylation in 293T and HCT116 cells. (G) CDK4, Cyclin-D, and CDKN1A-C: expression and interactions in 293T and HCT116 cells. (H) DREAM Complex interactions in 293T and HCT116 cells. (I) 293T-specific interactions involving proteins whose expression is regulated by the DREAM complex.

Figure 5:. Data-driven Discovery of Shared and Cell-line-specific Network Communities

(A) 293T and HCT116 networks were combined and partitioned via MCL clustering to identify 1,423 communities with 3+ members. Interactions connecting community pairs were then tallied to identify 1,736 statistically enriched associations. Interaction overlap across cell lines was then tallied within each community and along edges that connect associated community pairs. (B) Network of communities detected in the combined 293T/HCT116 network. Every network community with at least 3 members is represented as a node of size proportional to the number of proteins it contains. Edges connect communities that were statistically associated. Node and edge colors reflect overlap among cell lines. (C) Ternary diagram depicting the overlap observed within each community. Only communities for which at least one member has been a bait in both cell lines are included. (D) Ternary diagram depicting the overlap observed for edges that connect communities. Edges were only included if they were supported by 3+ edges detectable in both cell lines given the baits targeted in each. (E) – (G) Selected Network Communities.

Figure 6.. Cell Line Specificity among Domain Associations

(A) PFAM domains were mapped to proteins in the combined HCT116/293T network and domain pairs connected by unusually high edge counts identified. The overlap of edges connecting each statistically associated domain pair was then determined across cell lines. These domain associations form a network with edge colors that reflect sharing of interactions across cell lines. A ternary plot depicts sharing of edges matching each domain association across cell lines. The box-whisker plot shows the fraction of interactions shared among cell lines; a histogram highlights the fraction of domain associations that are shared or cell-line-specific. (B) – (E) Subnetworks of PFAM domain pairs. P-values reflect enrichment of interactions among the indicated domain pair with multiple testing correction. Only edges eligible for detection in both cell lines are shown.

Figure 7.. Linking Physical and Functional Associations for Biological Discovery

(A) For each interacting protein pair in the combined 293T/HCT116 network, cellular fitness profiles from Project Achilles were correlated and assessed for statistical significance. Following multiple testing correction, edges with positive or negative fitness correlations were extracted and assigned as either shared or cell-specific. Only edges detectable in both cell lines are shown. (B) – (C) BioPlex subnetworks with positive (B) or negative (C) fitness correlations. Edges connect proteins that interact in BioPlex and whose fitness profiles correlate (5% FDR). (D) – (F) Subnetworks of panel B. Green labels summarize common biological themes in each subnetwork. (G) – (I) Subnetworks of panel C. Green labels summarize common biological themes in each subnetwork.