Genomic Determinants of Protein Abundance Variation in Colorectal Cancer Cells

Abstract

Assessing the impact of genomic alterations on protein networks is fundamental in identifying the mechanisms that shape cancer heterogeneity. We have used isobaric labeling to characterize the proteomic landscapes of 50 colorectal cancer cell lines and to decipher the functional consequences of somatic genomic variants. The robust quantification of over 9,000 proteins and 11,000 phosphopeptides on average enabled the de novo construction of a functional protein correlation network, which ultimately exposed the collateral effects of mutations on protein complexes. CRISPR-cas9 deletion of key chromatin modifiers confirmed that the consequences of genomic alterations can propagate through protein interactions in a transcript-independent manner. Lastly, we leveraged the quantified proteome to perform unsupervised classification of the cell lines and to build predictive models of drug response in colorectal cancer. Overall, we provide a deep integrative view of the functional network and the molecular structure underlying the heterogeneity of colorectal cancer cells.

Keywords: CRISPR/cas9; TMT; cell lines; colorectal cancer; drug response; mutations; networks; phosphorylation; protein complexes; proteomics.

Copyright © 2017 The Author(s). Published by Elsevier Inc. All rights reserved.

Figures

Graphical abstract

Figure 1

Global Distributions of Gene-to-Gene Correlations and Protein Co-variation Networks in Colorectal Cancer Cell Lines (A) Distributions of Pearson’s correlation coefficients between protein-protein pairs (left panel) and mRNA-mRNA pairs (right panel) for all pairs (gray) and for pairs with known interactions in the CORUM database (blue). (B) Receiver operating characteristic (ROC) curves illustrating the performance of proteomics- and transcriptomics-based correlations to predict CORUM and high-confident STRING interactions. (C) Protein abundance correlation networks derived from WGCNA analysis for enriched CORUM complexes. The nodes are color-coded according to mRNA-to-protein Pearson correlation. (D) The global structure of the WGCNA network using modules with more than 50 nodes. Protein modules are color coded according to the WGCNA module default name, and representative enriched terms are used for the annotation of the network. See also Figure S3.

Figure 2

The Effect of Colorectal Cancer Driver Mutations on Protein Abundances (A) Association of driver mutations in colorectal cancer genes with the respective protein abundance levels (ANOVA test; permutation-based FDR 

Figure 3

The Global Effects of Genomic Alterations on Protein and mRNA Abundances (A) Volcano plot summarizing the effect of missense mutations on the respective protein abundances (ANOVA test). Hits at permutation-based FDR 

Figure 4

The Consequences of Mutations on Protein Complexes (A) Correlations networks filtered for known STRING interactions of proteins downregulated by LoF mutations at p value 10(p value). CORUM interactions are highlighted as green thick edges, and representative protein complexes are labeled. (B) Protein abundance correlation network of the ARID1A, ARID2, and PBRM1 modules. Green edges denote known CORUM interactions, and the edge thickness is increasing proportionally to the WGCNA interaction weight. (C) Heatmap summarizing the protein and mRNA abundance log2fold-change values in the knockout clones compared to the wild-type (WT) clones for the proteins in the ARID1A, ARID2, and PBRM1 modules. (D) Volcano plots highlighting the differentially regulated mRNAs in the KO samples. (E) Scatterplot illustrating the correlation between protein and mRNA abundance changes in the ARID1A KO. (F) KEGG pathway and CORUM enrichment analysis for the proteomic analysis results of ARID1A, ARID2, and PBRM1 CRISPR-cas9 knockouts in human iPSCs.

Figure 5

Proteome-wide Quantitative Trait Loci Analysis of Cancer Driver Genomic Alterations (A) Identification of cis and trans proteome-wide quantitative trait loci (pQTL) in colorectal cancer cell lines considering colorectal cancer driver variants. The p value and genomic coordinates for the most confident non-redundant protein-variant association tests are depicted in the Manhattan plot. (B) Replication rates between independently tested QTL for each phenotype pair using common sets of genes and variants (n = 6,456 genes). (C) Representation of pQTL as 2D plot of variants (x axis) and associated genes (y axis). Associations with q 

Figure 6

Proteomics Subtypes of Colorectal Cancer Cell Lines and Pathway Analysis (A) Cell lines are represented as columns, horizontally ordered by five color-coded proteomics consensus clusters and aligned with microsatellite instability (MSI), HNF4A protein abundance, cancer driver genomic alterations, and differentially regulated proteins. (B) KEGG pathway and kinase enrichment analysis per cell line. See also Figure S6.

Figure 7

Pharmacoproteomic Models (A) The number of drugs for which predictive models (i.e., models where the Pearson correlation between predicted and observed IC50s exceeds r > 0.4) could be fitted is stratified per data type. (B) Heatmap indicating for each drug and each data type whether a predictive model could be fitted. Most drugs were specifically predicted by one data type. (C) Heatmap of scaled log2 IC50 values for selected drugs displaying significant association (ANOVA FDR < 0.05) between protein abundance of ABCB1, ABCB11, and drug response. (D) Dose-response profiles for colorectal cancer cell lines treated with docetaxel (black line), 2.5 μM tariquidar alone (gray dotted line), or the combination of docetaxel and 2.5 μM tariquidar (orange line). Error bars represent mean ± SEM. See also Figure S7.