Bismark, bisulfite and chloroplast DNA

[colindaven]

Dear all,

I have done a bismark alignment of high coverage bisulfite Illumina 100bp PE reads (maybe 50-60x) to a plant genome. The biologists I work with are slightly unimpressed that Cs in the chloroplast are predicted to be methylated, since they should apparently remain unmethylated.

This might be due to
1) incomplete bisulfite treatment
2) mismapped genomic DNA reads mapped to the chloroplast (for example a mapping quality problem)
3) contamination of genomic DNA by the many chloroplasts present in the leaf material used.
4) further ideas?

Which of these do you think might be more probable ?

Regards,
Colin

[fkrueger]

Hi Colin,

I guess you can narrow it down somewhat by looking at the data in some more detail. For example, what are the methylation rates in the 3 contexts, especially in CHH context? Do the overall numbers fit with what is known about your organism?
I haven't worked with plant methylation before but I suppose methylation in CHH context should be the lowest methylation overall. If this value is low you can probably already get an idea about the bisulfite conversion rate error you have in your library (i.e. if the CHH methylation is 2% for example you have to assume that the conversion efficiency must have been > 98%. Does the methylation in chloroplasts exceed this rate?

Next you could import the data into a genome browser and look at the read distribution. This might tell you if large stretches of the chloroplast are completely unmethylated and if there are small patches that appear fully methylated. This scenario would be in favour of mapping artifacts, and in fact I have seen just that for some mitochondrial methylation in some mouse data recently. Alternatively, you might see a low average methylation across the entire chloroplast, which would argue against mismapping events. You could either have a few reads that are fully methylated (possible contaminants after the bisulfite conversion?) and many unmethylated reads, or just a mix of methylated and unmethylated cytosines within the two within the same reads. Importing the reads into SeqMonk should enable you to very quickly answer some if not all of these questions.

Just as a thought, there also might be some methylation in the chloroplasts ?!

[colindaven]

Hi Felix,

thanks for your prompt answer and tips. We have been testing JBrowse.

We're talking non-model organism here. Happily, we screened the mt and chloroplast for errors and found some very poorly assembled sections and some genomic contamination. The mt methylation was centred around a region from a falsely integrated, possibly genomic section.

I am redoing the alignment with better chloroplast and mitochondrion as we speak.

To come back to mapping quality though, does Bismark give any mapping quality parameters out, or should the user investigate the SAM file directly ?

Interestingly, I had an error with the most recent version of bowtie2 v2.2.3, which was corrected by going back to previous installed version 2.1.0. This might be a good one to test at your end when you have a moment.

No index, query, or output file specified!
Bowtie 2 version 2.2.3

[fkrueger]

Hi Felix,

thanks for your prompt answer and tips. We have been testing JBrowse.

We're talking non-model organism here. Happily, we screened the mt and chloroplast for errors and found some very poorly assembled sections and some genomic contamination. The mt methylation was centred around a region from a falsely integrated, possibly genomic section.

I am redoing the alignment with better chloroplast and mitochondrion as we speak.

To come back to mapping quality though, does Bismark give any mapping quality parameters out, or should the user investigate the SAM file directly ?

Interestingly, I had an error with the most recent version of bowtie2 v2.2.3, which was corrected by going back to previous installed version 2.1.0. This might be a good one to test at your end when you have a moment.

No index, query, or output file specified!
Bowtie 2 version 2.2.3

Hi Colin,

Good to see that you already spotted some potential sources for the behavior you saw. With Bismark version 0.12.1 we added the calculation of MAPQ scores to the BAM output, if you wanted to use that to assess mapping quality. The most recent posts of the main Bismark thread also contain some discussion about filtering based on MAPQ scores.

Regarding the Bowtie 2 error: I remember that the Bowtie2 command withing Bismark used to work until a few Bowtie 2 releases ago, but then they fixed something that was a little bug in Bowtie 2 which broke the command in Bismark. This has been fixed in Bismark, so if you would run the latest version of Bismark it should work (which version were you running?). Please find attached the latest version of Bismark attached, it is even a bit newer than the latest release on the website. We are currently running Bowtie 2 version 2.2.2 on our cluster here and it all works fine. I shall test version 2.2.3 shorty, but I need to make some preparations for the France - Germany match, too...

Please let me know if the latest Bismark version works for you, and sorry for the confusion with all the version numbers... Cheers, Felix

[fkrueger]

I just tested Bowtie 2 version 2.2.3 and it seems to work fine as well.

[colindaven]

Hi Felix,

yes, Bowtie2 works fine with the more up to date versions of Bismark. I was running 0.11.1 I believe.

Thanks for your advice on mapping quality.

Colin