Binary methylation states after bisulfite sequencing

[maf5]

Hi everyone. I have sequencing data (illumina Hiseq) from bisulfite treated samples (Arabidopsis). The bisulfite conversion of some selected regions was tested by sanger sequencing initially and looked fine. The library was then made from the same samples using the Diagenode microplex kit.

Here's the problem: almost all of the cytosines show either total, or zero methylation levels. There are very few (<1%) that show any intermediate levels of methylation. Even on the regions that were already tested by sanger sequencing, which showed a complete range of methylation levels from 0 to 100%.

So, does anyone know how this bias can come about. My suspicion is that it is either due to an amplification bias when making the libraries, or something in the bioinformatics pipeline. Has anyone experienced similar results from bisulfite sequencing?

Cheers

[dpryan]

How were the samples mapped? I can't really debug issues with the library prep (I'm sure someone else here can), but at least the downstream stuff I can offer advice on.

[nucacidhunter]

Hi everyone. I have sequencing data (illumina Hiseq) from bisulfite treated samples (Arabidopsis). The bisulfite conversion of some selected regions was tested by sanger sequencing initially and looked fine. The library was then made from the same samples using the Diagenode microplex kit.
Cheers

I assume you are referring to whole genome bisulfite sequencing and wonder how you have made a library with bisulfite converted DNA. The kit input is dsDNA while BS converted DNA will be single stranded.

[maf5]

Thanks for the replies, it's taken a little time for me to get the details from our bioinformatician.

The library was pair-end sequenced. To increase the coverage, I mapped each of the mates as independent libraries. I used bowtie2 to map against the in-silico bisulfite-converted TAIR10 genome. I use a seed length of 25 and allowed at most one mismatch in the seed. If a read maps to multiple positions, I keep only the best alignment in terms of quality and numbers of mismatches.
Using the mapping information I then remove duplicates.

Regarding the library preparation, the bisulfite treatment was performed on non-denatured DNA so the majority of the genome would have actually remained double stranded (we are not actually looking for methylation in this case) but the library prep worked fine with the diagenode kit, although with quite a few amplification rounds, which is why I suspect it may be an amplification bias

[GenoMax]

The library was pair-end sequenced. To increase the coverage, I mapped each of the mates as independent libraries.

How is that going to increase coverage?

[dpryan]

The bioinformatician must be rather new to things. Please advise him/her to not try needlessly reinventing the wheel and instead use bwa-meth or bismark or one of the many preexisting alignment programs. It's incredibly likely that he/she simply screwed things up.

Edit: Forgot an important "not" above!

[ecSeq Bioinformatics]

DNA Methylation Data Analysis Workshop
How to use bisulfite-treated sequencing to study DNA methylation

When?
22. - 25. November 2016

Where?
Leipzig, Germany

Link?

http://www.ecseq.com/workshops/workshop_2016-07-NGS-DNA-Methylation-Data-Analysis