Nearby Cpg Islands Methylation And Gene Expression

Question

[qliu2011]

I'm currently working on a research project on finding the effect of methylation and gene expression. Because the human genome is so large and I only have limited amount of WGBS sequencing data for different epigenomes and Roadmap gene expression data. I want to ask biologically (without counting promoter, I will take promoters into account separately), is it beneficial to just consider the nearest intergenic CpG island methylation in looking at the expression of a gene? If not, what other factors should I consider? Thanks.

[Sapna]

I'm currently doing my PhD, and have taken epigenetics as my research topic. I am supposed to identify certain candidate genes under drought. I have further analyzed the promoter regions for any CpG islands. and based on that I have selected some genes (DEGs). All in silico analysis using different microarray datasets (and not epigenome datsets). I have done this anlaysis in rice. Would like to know if the method I am following is correct. Or need to modify certain things?

Answer 1

[Charles Warden]

It may be hard to clearly distinguish between promoter and CpG island, but I would say "Yes".

In fact, the COHCAP algorithm integrates gene expression and DNA methylation data (Illumina array or targeted BS-Seq) using nearby CpG islands:

http://www.ncbi.nlm.nih.gov/pubmed/23598999

[brentp]

I like how you have the explicit relation between methylation and expression--more software should do this. (But) We know that the text-book relation between methylation and expression is inverse (increase methylation -> decreased expression), but in practice, it is often increased expression -> increased methylation--which, if I understand correctly, your software will ignore. Thoughts on those? I'm sure you've seen them as it seems that the proposed inverse relationship is the exception rather than the norm.

[Charles Warden]

Actually, I have found the UCSC CpG Islands to typically follow canonical rules (especially if they are in an appropriate location, such as the promoter region).

The algorithm is designed to narrow down genes with a negative correlation in methylation and gene expression (since that is typically desired for validation purposes), but I think Supplemental Figure S13 addresses your question: you can see that negative correlations are much more common than positive correlations for CpG islands and nearby genes. Maybe you would also be interested in Figure S12, but I think that one is less important.