Whole-genome analysis of DNA methylation across cancer types reveals specific patterns in early stages

• Main Author: Mestre, André Miguel Romeira
• Publication Date: 2019
• Format: Master thesis
• Language: eng
• Source: Repositórios Científicos de Acesso Aberto de Portugal (RCAAP)
• Download full: http://hdl.handle.net/10400.1/13567
• topic: Cancro; Metilação de DNA; Expressão genética; Biomarcador de diagnóstico e análise computacional

Dynamic variations in DNA methylation are known to play an important role in cancer development through modulation of gene expression. Here, were developed a mathematical structured model to identify patterns of differentially methylated genes (cDMGs), across different cancers types that can act as epigenetic diagnostic biomarkers. A Working Pipeline (WP), designed in R language, was applied to 8 cancer cohorts from The Cancer Genome Atlas (TCGA) aiming to analyze DNA methylation and gene expression alterations occurring during normal to stage I carcinogenic transition. WP has a principal component which was divided in four steps: 0. Clinical characterization of patients; 1. Identification of cDMGs; 2. Identification of genetic/epigenetic patterns across different cancer type; and 3. Identification of diagnostic predictors. Additionally, the WP had a second component containing two more complementary steps: 4. Identification of CpG probes that better predict gene expression and 5. HJ-Biplot approach to visualize genes or CpG probes and its association with sample distribution. Appling the principal component of the WP to TCGA cohorts, we identified 117 cDMGs in breast cancer, 307 in colorectal cancer, 99 in head and neck cancer, 156 in kidney clear cell cancer, 106 in kidney papillary cancer, 349 in liver cancer, 180 in lung cancer and 25 in thyroid cancer. Analysis of patterns across these cancers revealed that the majority of cDMGs are cancer-specific. Moreover, we found cDMGs to be good predictors of diagnosis. When considering specific biomarkers for each cancer, only 19, 153, 27, 93, 53, 72, 38 and 14 genes were found to be good diagnostic biomarkers in breast, colorectal, head and neck, kidneyR, kidneyP, liver, lung and thyroid cancers, respectively. Therefore, we developed a novel working pipeline that allowed data sets analyses available worldwide. Validation of this mathematical model evidences that normal-tumor transition is not a conserved process event across different cancers type, but specific to the cell of origin.