How to estimate representation of RRBS library

[liaojinyue]

Dear all,

We have recently generated RRBS libraries and sequenced them. Although the CpG site coverage is acceptable (~1M with coverage > 10), we are puzzled by the alignment result. Enclosed figure shows the alignment results from two of our samples covering one fragment with MspI sites at both ends. However, we only detect reads deriving from positive strand for the first sample (coverage = 186 reads) and reads from negative strand for the second sample (coverage = 58reads). My understanding is that the positive and negative strand should be more or less equally amplified and we should be able to detect reads from both strands in each sample. The second figure shows the methylation value density for the CpG sites. It looks like the methylation levels are either 0% or 100%. Our concern is that reads mapped to each MspI digested fragment might all come from one template. Does it mean that we do not have sufficient representation due to sample loss during library preparation? I will be grateful for your help.


Best,

Jason

[dpryan]

In my experience this sort of thing ends up being fairly common in RRBS. My guess is that PCR amplification leads to this sort of bias.

[nucacidhunter]

All current RRBS methods cover both strands unless you have used a novel method that targets only one strand. You have not mentioned if observed strand specificity in libraries is for all fragments or just for a subset of them. If your data analysis is correct and all fragments in a library come from a particular strand, I would suspect that library prep has not been optimal.

If your adapters did not include UMIs it will be difficult to know if reads for a fragments are PCR duplicates or originate from different restriction fragments.

[ahlinyaobatherbal]

Why my backlink can not be read ?

[Diagenode]

I think we are dealing with two issues here.
The first one is the question of extreme values (all 0 or 100% methylation), which I think is normal, or at least usual in RRBS-seq. Especially if you work with cell lines, the cells have no reason to have different methylation profiles, but even if you prepare biopsies from a fairly homogene tissue like liver, I would not expect a lot of cell-to-cell variation, which means that a certain CpG site is either methylated or unmethylated in all cells uniformly, so the sequencing results should reflect this. We have recently done RRBS-seq on a batch of samples, I quickly checked the methylation ratios of CpGs, and I also noticed that over 90% of all the covered CpGs are methylated either >90% or <10%.
And remember that RRBS mostly capture the most prominent CpG islands, that have a crucial role in gene expression regulation. Other CpGs, like the ones in CpG shores and shelves and further away from the islands are less represented, especially if you have a low coverage, and I think it is these CpGs where we can expect a bigger variation, as they are likely to have a less essential role. There are also studies that prove that C methylation plays a role in other context than CpG, eg. CHG (see for example http://nar.oxfordjournals.org/content/42/5/3009.full) in these contexts probably there are more variations too. But CpG islands seem to be quite consistent (within a tissue or a cell line at least).

The other issue (getting reads only from a single strand) is far from normal. It must be some library preparation issue; as nucacidhunter said, both strands should be covered more or less equally. I can only recommend this kit: Premium RRBS, which we use routinely for our RRBS-seq service, and all of our customers are very satisfied with it so far, we haven't observed the strand bias you described. It generates a pool of several samples for a cost-effective sequencing, and it's very user friendly, easy to use, difficult to make mistakes, not to mention the great CpG representation and coverage we get.

[Niels Wagemaker]

We had some bias in the watson crick # reads but when we starter doing epigbs with 5mc insensitive restriction enzymen this was solved...