Differential methylation analysis: sequencing biological replicates versus pooled samples?

Question

[Ian]

I have been discussing a future differential methylation analysis (DMA) using RRBS, but I cannot decide whether it is best to recommend multiple replicates per condition, or one sample per condition containing pooled mouse samples.

The biological sample is from a specific part of mouse brain so the DNA yield, per sample, is going to be low. I usually use Methylkit for DMA, which can use either single sample comparison or replicates.

So in this case would it be acceptable/best to pool tissue from multiple mice into one sample, which would ensure an adequate DNA yield for BS treatment and library prep, OR are multiple replicates best for DMA?

Thank you for any opinions!

[Jean-Karim Heriche]

In my experience analyzing biological data, it is always better to have replicates. You want to be able to assign any variation you observe in your results to the conditions you're comparing and avoid confounding variation from technical or biological origin. To control for these, you need replicates. Also low amount of material usually give results with high variance so it seems you should probably either pool mice or do even more replicates. If access to mice is not an issue, I don't see why you couldn't get replicates of pooled samples.

Answer 1

[Shicheng Guo]

Different study design or sequencing strategy would provide you different information. Suppose the heterogeneity of the pooled samples are very high, then it would be inappropriate to pooled them. In this situation, such operation of pooling would decrease the differential detection power of RRBS (Because the variation is high within the group). Therefore, before you make such decision, you need to know the variation among these samples. The variation can be estimated or known from previous report. On the other side, when the heterogeneity of the samples is lower, sure, you can pool them and consider them as a biological replication. In our previous study, we used pooled library to identify cancer abnormal methylation regions with MBD-seq and with this strategy (10 samples per library and we will prepare 5 or 6 library for cancer and normal groups), we can identify some differential methylation regions. In conclusion, no matter, pooling or non-pooling were choose, you need replications for each group/subsets.

Answer 2

[Devon Ryan]

If they really need to pool in order to have sufficient DNA (I've extracted plenty of brain regions and never had that issue, but perhaps these are prenatal brains) then just have them pool within a litter for each sample. For mice, biological replicates must be different litters anyway, so they won't lose too much by pooling. In general, though, I agree with Jean-Karim Heriche and would suggest that they sequence individual mice with sufficient number of replicates if there's sufficient material.