Tissue-Specific Signaling Networks Rewired by Major Somatic Mutations in Human Cancer Revealed by Proteome-Wide Discovery

Abstract

Massive somatic mutations discovered by large cancer genome sequencing projects provide unprecedented opportunities in the development of precision oncology. However, deep understanding of functional consequences of somatic mutations and identifying actionable mutations and the related drug responses currently remain formidable challenges. Dysfunction of protein posttranslational modification plays critical roles in tumorigenesis and drug responses. In this study, we proposed a novel computational oncoproteomics approach, named kinome-wide network module for cancer pharmacogenomics (KNMPx), for identifying actionable mutations that rewired signaling networks and further characterized tumorigenesis and anticancer drug responses. Specifically, we integrated 746,631 missense mutations in 4,997 tumor samples across 16 major cancer types/subtypes from The Cancer Genome Atlas into over 170,000 carefully curated nonredundant phosphorylation sites covering 18,610 proteins. We found 47 mutated proteins (e.g., ERBB2, TP53, and CTNNB1) that had enriched missense mutations at their phosphorylation sites in pan-cancer analysis. In addition, tissue-specific kinase-substrate interaction modules altered by somatic mutations identified by KNMPx were significantly associated with patient survival. We further reported a kinome-wide landscape of pharmacogenomic interactions by incorporating somatic mutation-rewired signaling networks in 1,001 cancer cell lines via KNMPx. Interestingly, we found that cell lines could highly reproduce oncogenic phosphorylation site mutations identified in primary tumors, supporting the confidence in their associations with sensitivity/resistance of inhibitors targeting EGF, MAPK, PI3K, mTOR, and Wnt signaling pathways. In summary, our KNMPx approach is powerful for identifying oncogenic alterations via rewiring phosphorylation-related signaling networks and drug sensitivity/resistance in the era of precision oncology. Cancer Res; 77(11); 2810-21. ©2017 AACR.

Figure 1. Workflow of kinome-wide network module identification for cancer pharmacogenomics

(A) Somatic mutation profiles, kinase phosphorylation spectrum, tissue-specific co-expression network, and drug response data were integrated for kinome-wide network module search in cancer pharmacogenomics. (B) Diagram of the algorithm to detect the consensus mutant kinase-substrate network modules. (C) Identifying new cancer proteins or network module genes harboring the enriched somatic mutations at their phosphorylation sites that are also associated with patient survival and drug responses.

Figure 2. Mutation frequencies and distribution patterns in 16 cancer types

(A) Missense mutation frequencies across 16 major cancer types. Distribution of missense mutation frequencies (number of mutations per 1 million residues) in whole protein sequences (All) versus phosphorylation sites in 16 major cancer types. (B, C) Cumulative distribution of deleterious mutations at direct phosphorylation sites, their seven immediate flanking residues, and outside positions. Cumulative frequencies of SIFT (B) and PolyPhen-2 scores (C) for direct phosphorylation sites (abbreviated as D), +/−7 flanking residues (D + 1–7 flanking positions), and outside positions. The 16 major cancer types are acute myeloid leukemia (LAML), bladder urothelial carcinoma (BLCA), breast invasive carcinoma (BRCA), colon and rectal adenocarcinoma (COAD/READ), glioblastoma multiforme (GBM), head and neck squamous cell carcinoma (HNSC), kidney chromophobe (KICH), kidney renal clear cell carcinoma (KIRC), lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), ovarian serous cystadenocarcinoma (OV), prostate adenocarcinoma (PRAD), skin cutaneous melanoma (SKCM), stomach adenocarcinoma (STAD), thyroid carcinoma (THCA), and uterine corpus endometrial carcinoma (UCEC).

Figure 3. Manhattan plot of putative significantly mutated proteins identified in pan-cancer analysis (A) and individual cancer analysis (4 selected individual cancer types) (B)

Each dot represents a gene or its protein. Red dots represent the Cancer Gene Census (CGC) genes. Yellow dots represent the significantly mutated genes collected from literatures (Table S2). The horizontal red line denotes the false discovery rate at 0.05. The putative significantly mutated proteins identified in the remaining 12 cancer types were showed in Figure S1. Abbreviations of 16 cancer types in Figures 3 and 5 are provided in the legend of Figure 2.

Figure 4. The consensus mutant kinase-substrate network module identified for uterine corpus endometrial carcinoma (UCEC)

(A) The subnetwork for 11 genes in the final consensus module for UCEC. The size of nodes denotes the frequencies of genes identified in individual network modules quantified by the objective score (Figure 1B). (B) The mutual exclusivity of the phosphorylation site mutation profiles for the 11 genes in UCEC. (C) The three-dimensional model of phospho-degron motif of β-catenin (encoded by gene CTNNB1). PDB: 1P22 for β-TrCP1-Skp1-β-catenin complex was downloaded from PDB database (http://www.rcsb.org/). White-black: Skp1; purple: β-catenin phospho-degron motif; and the remaining color: β-TrCP1. Two phosphorylation sites: p.S33 and p.S37 were highlighted in red. (D) Distribution of missense mutations at p.S37 site in the 66 UCEC samples as shown in Figure 4B.

Figure 5. Representation of phosphorylation site mutations from primary tumors in cancer cell lines across 15 cancer types

Pearson correlation of phosphorylation site mutation frequency between cell lines and primary tumors for each cancer type. The number in parentheses refers to the sample size in each cancer type.

Figure 6. Kinome-wide landscape of pharmacogenomic interactions

(A) Pan-cancer ANOVA analysis for statistically significant interactions between phosphorylation site mutations and differential drug sensitivity. (B) Examples of significant pharmacogenomic interactions for mutant (Mut) vs. wild-type (WT) cell lines identified by ANOVA. The details for the remaining significant pharmacogenomic interactions identified by ANOVA are provided in Table S5. The color keys in A represent the P-value (-Log10) of ANOVA analysis for the drug sensitivity (red) or drug resistance (blue) caused by phosphorylation site mutations in human cancer cell lines (see Methods).

Figure 7. A proposed model to illustrate the potential molecular mechanisms of kinome-wide landscape of pharmacogenomic interactions altered by phosphorylation site mutations as shown in Figure 6A

The FDA-approved or clinical investigational cancer drugs targeting 5 selected signaling pathways, including PI3K signaling, MAPK signaling, EFG signaling, mTOR signaling, and Wnt signaling, were illustrated. The data was provided in Figure 6 and Table S5.