Bottom-up, integrated -omics analysis identifies broadly dosage-sensitive genes in breast cancer samples from TCGA

Abstract

The massive genomic data from The Cancer Genome Atlas (TCGA), including proteomics data from Clinical Proteomic Tumor Analysis Consortium (CPTAC), provides a unique opportunity to study cancer systematically. While most observations are made from a single type of genomics data, we apply big data analytics and systems biology approaches by simultaneously analyzing DNA amplification, mRNA and protein abundance. Using multiple genomic profiles, we have discovered widespread dosage compensation for the extensive aneuploidy observed in TCGA breast cancer samples. We do identify 11 genes that show strong correlation across all features (DNA/mRNA/protein) analogous to that of the well-known oncogene HER2 (ERBB2). These genes are generally less well-characterized regarding their role in cancer and we advocate their further study. We also discover that shRNA knockdown of these genes has an impact on cancer cell growth, suggesting a vulnerability that could be used for cancer therapy. Our study shows the advantages of systematic big data methodologies and also provides future research directions.

Fig 1. Bottom-up, integrated analysis workflow.

Visual representation of the analytical workflow for identifying broadly dosage-sensitive genes. Green portions represent mRNA-based steps, red protein, blue CNV. Integrated and filtering steps are white. Briefly, data were acquired from their sources, joined with metadata, normalized, integrated, then filtered.

Fig 2. mRNA fold change versus protein fold change.

This represents the matched protein (y-axis) and mRNA (x-axis) z-score normalized, log₂ fold changes of disease to healthy samples for 106 patients and 20531 genes. Poor correlation amongst the points suggests that mRNA changes do not necessarily predict protein production changes.

Fig 3. Protein vs mRNA fold changes with CNV amplification.

Figures represent the z-score normalized, log2 fold change of mRNA (x-axis) versus protein (y-axis) and are colored by CNV segment mean values for samples from patients in the TCGA Breast Cancer dataset selected for: (a) Genes in the COSMIC database labelled as amplified, (b) known oncogene MYC, (c) known oncogene and BDSG HER2/ERBB2, and (d) all BDSGs. The trend in Fig 3A exemplifies how oncogenecity does not always correlate with dosage-sensitivity.

Fig 4. Heatmap and hierarchical clustering of shRNA knockdown.

Section (a) represents all cell lines available in the Achilles project and (b) breast-specific cell lines. Rows are the selected genes: blue are housekeeping genes, red are oncogenes, green oncosupressors, and light blue are BDSGs; columns are the cell lines. Red cell values represent cellular proliferation, blue cellular death, and white no change. The clustering of ERBB2 and two BDSGs (GRB7, RPS6KB1) in (b) suggests an oncogenic role in breast cancer. PPME1 and UBE2Z signatures in (a) and (b) suggest an overall oncosuppressive role.