Quantitative network mapping of the human kinome interactome reveals new clues for rational kinase inhibitor discovery and individualized cancer therapy

Abstract

The human kinome is gaining importance through its promising cancer therapeutic targets, yet no general model to address the kinase inhibitor resistance has emerged. Here, we constructed a systems biology-based framework to catalogue the human kinome, including 538 kinase genes, in the broader context of the human interactome. Specifically, we constructed three networks: a kinase-substrate interaction network containing 7,346 pairs connecting 379 kinases to 36,576 phosphorylation sites in 1,961 substrates, a protein-protein interaction network (PPIN) containing 92,699 pairs, and an atomic resolution PPIN containing 4,278 pairs. We identified the conserved regulatory phosphorylation motifs (e.g., Ser/Thr-Pro) using a sequence logo analysis. We found the typical anticancer target selection strategy that uses network hubs as drug targets, might lead to a high adverse drug reaction risk. Furthermore, we found the distinct network centrality of kinases creates a high anticancer drug resistance risk by feedback or crosstalk mechanisms within cellular networks. This notion is supported by the systematic network and pathway analyses that anticancer drug resistance genes are significantly enriched as hubs and heavily participate in multiple signaling pathways. Collectively, this comprehensive human kinome interactome map sheds light on anticancer drug resistance mechanisms and provides an innovative resource for rational kinase inhibitor design.

Figure 1. Diagram of systems biology-based framework for the human kinome interactome map building

This human kinome interactome map across 538 kinase genes includes five components: (i) kinase-substrate interaction network, (ii) physical protein-protein interaction network (PPIN) and an atomic resolution three-dimensional structural PPIN, (iii) drug-target interaction network, (iv) disease gene annotations, and (v) network, pathways, and bioinformatics analyses.

Figure 2. Functional annotations of the human kinome

(A) Pie chart of 538 kinase genes grouped by 10 different kinase groups: tyrosine kinases (TK), tyrosine kinase-like kinases (TKL), casein kinases (CK1), PKA/PKG/PKC-family kinases (AGC), calcium/calmodulin-dependent kinases (CAMK), sterile homologue kinases (STE), CDK/MAPK/GSK3/CLK-family kinases (CMGC), receptor guanylate cyclases (RGC), atypical protein kinases (Atypical), and kinases that did not belong to any groups above (Other). (B) The Venn diagram of overlaps among 538 kinase genes, 1,855 drug target proteins, 487 Cancer Gene Census (CGC) genes, and 2,721 essential genes. (C) The Venn diagram of overlaps among 538 kinase genes, 2,714 Mendelian disease genes (MDGs), 2,123 orphan disease-causing mutant genes (ODMGs), and 2,721 essential genes.

Figure 3. Kinase-substrate interaction network (KSIN)

The size of each node reflects its degree of connectivity in KSIN. Abbreviations of kinase groups (circles) are provided in the Figure 2 legend. Non-kinase substrate nodes (squares) are color-coded according to their phosphorylation sites, including phosphoserine (pS), phosphothreonine (pT), and phosphotyrosine (pY).

Figure 4. Sequence motif analysis of kinase phosphorylation sites

(A) Logo analysis of target phosphorylation site sequence motifs (four amino acids before and after the phosphorylation residues) for 12 kinases that have the strongest connectivity in kinase-substrate interaction network. The amino acids are labeled according to their chemical properties: green for polar amino acids (G, S, T, Y, C, Q, N), blue for basic amino acids (K, R, H), red for acidic amino acids (D, E), and black for hydrophobic amino acids (A, V, L, I, P, W, F, M). (B) A substrate peptide binding pocket of CDK2 (PDB ID: 1QMZ). (C) Another substrate peptide binding pocket of CDK2 (PDB ID: 1GY3). B and C were prepared using the software PyMOL (http://www.pymol.org/).

Figure 5. Kinase-drug interaction network

In this network, a drug node (square) and a target kinase node (circle) are connected to each other by a grey edge if the target is annotated as a known interaction with the drug. The size of each node reflects its degree of connectivity. Drug nodes (circles) are green (experimental drugs) or gold (FDA approved drugs). Kinase nodes (circles) are color-coded according to the kinase groups (see Figure 2 legend).

Figure 6. Network analysis of kinase inhibitor response

(A) Drug sensitivity network of 11 molecularly targeted kinase inhibitors (Supplementary Figure S5). This network includes four types of edges: kinase-drug interaction (gold solid line), drug-cancer association (red solid line), gene-drug sensitivity associations (purple solid line with arrow), and target gene-drug sensitivity associations (blue solid line with arrow). Color codes of nodes: drug (gold square), target gene or target protein (green circle), drug sensitivity genes (cyan-blue circle), drug target and sensitivity gene (red circle), and cancer (purple hexagon). (B) Volcano plot of sensitivity response to Gefitinib, an epidermal growth factor receptor (EGFR) inhibitor. The calculation of a p-value for each drug-gene association was described in a previous work [35]. The data was from the Genomics of Drug Sensitivity (http://www.cancerrxgene.org). (C) The simplified EGFR signaling pathways involving Gefitinib sensitivity through the RAS/MEK/ERK and PI3K/PDK1/AKT pathways.