RADseq library Prep- Newer protocols for pooled samples?

[Mwers1]

Hello All,

I am hoping to use RAD-seq for the discovery of SNPs for a population genomics investigation. I have a few questions related to library prep of pooled samples.

My main overarching question is: How can commercial library prep kits available from vendors be applied to creating pooled libraries?

Especially since my goal is to barcode individuals in within libraries?

Are updates available to protocols such as Etter et al., (2011) in light of the new products available from vendors?

Many thanks!

[SNPsaurus]

If the goal is just to discover SNPs, I would just Nextera prep 20 individuals and sequence them to 40X read depth. Assemble the pooled reads into a draft genome, align the reads to the genome and call a SNP whenever you see an alternate allele in 4 or more reads at a locus. How big is your genome? Even a human-sized genome will fit into a HiSeq lane, so the total cost of the project would max out at $3000 (a lane of paired-end 150 bp reads) unless you work on an absurdly-genomed plant or amphibian. You could do the same with RAD, and still need to spend the money for a lane of sequencing, but get many fewer SNPs and no draft reference. If there is already a reference, then you could use a lane of single-end reads and cut costs (or do that if you don't have a reference and assemble a worse reference).

[Mwers1]

Hi SNpsaurus!

Thanks for your reply. It is a bit more complicated as I am doing multiple populations and performing a reverse ecology study and looking for signatures of selection. In my present design, I am planning to sequence ~288 individuals lines in pooled libraries of 12.

My model, Daphnia, has a reference assembly so 200 Mb, I think. Currently available capabilities is a MiSeq system in Uni's Core.

Are you suggesting nextera prep all 20 individuals into a single library?

[SNPsaurus]

It sounds like you want to do more than SNP discovery, so my previous suggestion wouldn't be very good. But I would still suggest blasting away with whole genome sequencing for two reasons (and I say this as a developer of the original RAD-Seq protocol and a service provider for a different RAD method that we call nextRAD).

1) You want to find signatures of selection, so SNP density matters.

2) The genome is tiny.

I think individually tracking the samples can be helpful. In a pool you may see shifts of allele frequencies that are purely due to biased representation of an individual in the pool. With individual labelling you can also toss an individual from the data if you learn it looks weird or mis-labelled. A pool is harder to refine.

We would do a project of 288 Daphnia with individual library prep and sequencing to 1X read depth with paired-end 150 bp reads across the whole genome for each sample, including variant calling after alignment to the reference for $35/sample. You could probably do pooled RAD for less (maybe), but with much less power to detect selection and no ability to re-pool samples later based on new information. You could also do the whole genome sequencing yourself for about the same cost.

[Mwers1]

Hi!
I wonder if RAD would still be less powerful given some of the unique aspects of the daphnia model.

Particularly, I have temporal subpopulations spanning ~80+ years. So I have individuals from the population through the entire time for which the stressor has been applied to the population. Would this, do you think, mitigate some of the limitations of RAD's lower SNP count?

I'm stuck because I have very limited resources to do this project.

Thanks!

[SNPsaurus]

It could help since a randomly biased pool is unlikely to be biased over and over so you could find SNPs where the shift in allele frequencies is consistent over time. On the other hand, seeing multiple SNPs in a region behave the same way will give confidence, and you get that from WGS.

Pooling rather than individually barcoding will save some money. Maybe take most samples from the start and end time points and cut the number of samples? If you can detect selection with those pools then you might be able to get more funds to track the changes over time, since you've shown it can work. If you don't detect it...probably sequencing more samples from in-between times won't help.