We aim to develop statistical methods and software for analysis of DNA methylation data from human colorectal cancer samples [CRCs] and matched normal tissue. We hypothesize that because epigenetics determines mammalian cell phenotypes, it will be possible to reconstruct the phenotype of the tumor and its founder cell by comparing epigenomes sampled from opposite sides of the same CRC. Therefore, we will rank genes or cell pathways according to the degree of conservation of methylation status they exhibit, indicating which are likely to be most important during tumor growth. We propose three innovations to accomplish these goals. First, we exploit a new Illumina microarray which can measure methylation at ~850,000 CpG sites, allowing broad coverage of most human genes and enhancers. We will develop methods to enable us to conduct an analysis of such data. We will demonstrate this using a test dataset in which we have collected multi-regional sampling data (i.e., data in which we sample from a number of different parts of the same tumor) for 26 human colorectal tumors, along with paired samples of normal tissue for 6 of those patients and 9 other colons. Second, we propose a two-pronged attack designed to assess whether each CpG site should be classified as ?stable? or ?unstable? with respect to the degree of CpG variation permitted there.
In Aim 1 we propose methods that are purely statistical in nature;
In Aim 2 we propose methods that will be built upon an explicit mathematical model for tumor evolution. We will compare and contrast their results. An additional advantage of the second approach is that it will also allow us to reconstruct the epigenome of the founder cell. Third, we will assess conservation of variation within genes or pathways to assess which are most important during growth---pathways with smaller methylation differences between tumor sides are likely to be more important and under selective pressures. The significance of the proposed studies is that we will develop new methods to extract epigenetic information from multi-regional tumor sampling. Such data are rare at the moment, but will soon be routinely collected. For that reason, in our fourth aim we propose to produce and freely distributed software and Shiny applications. Our long-term goal is to facilitate more personalized and effective therapies that specifically target pathways or genes most important to the growth of individual CRCs. The development of methods and software to characterize variation in methylation patterns from multi-regional tumor sampling, and relate that to genes/pathways will facilitate this process, and the relative ease of obtaining epigenetic information using methylation arrays should allow widespread translation to other tumor types.
We aim to develop statistical methods and software for analysis of DNA methylation data from human colorectal cancer samples [CRCs] and matched normal tissue. We will rank genes or cell pathways according to the degree of conservation of methylation status they exhibit, thereby indicating which are likely to be most important during tumor growth. Our goal is to facilitate more personalized and effective therapies that specifically target pathways or genes most important to the growth of individual CRCs.
