Alternatives to Pippin Prep for ddRAD?

[tugecko]

My lab is interested in doing population connectivity/phylogeography on herps, and think that ddRAD is probably the best NGS option for us (since we don't need as many loci as for other projects, and we would rather have more overlap between samples). However, the Pippin Prep is a bit over our price range. I know that part of the point of ddRAD is that the size selection window is very precise, which is where the Pippin Prep comes in, but has anyone had success doing ddRAD with bead or manual gel size selection? Or somehow lowering the cost of the Pippin Prep, maybe by making their own gels?

Alternatively, would anyone suggest a different RAD method for our analyses (we've also been considering RAD-PE, from Etter et al. 2011)?

Just starting out with NGS, any advice would be appreciated!

[Carcharodon]

Hey tugecko,

I'm doing ddRAD as well. Size-selection with beads tends to leave you with a relatively wide range of fragment sizes (folks in my lab do ezRAD quite a bit, and this is the method they most commonly choose). If a tight size-selection is what you're aiming for, I might avoid using beads. And SPRI beads are deceptively expensive in their own right.

Also, I haven't heard of anybody making their own gels for use in the Pippin-Prep. I don't know if you've see one, but it's a pretty complex little cartridge! I don't really think it can be done, to be honest.

As for the manual gel excision, I can't really say I've tried it for ddRAD. But the tighter your range, the greater the risk for allele drop-out. Based on the gels I've run (PCR, PCR product excision, etc.), I wouldn't be inclined to trust it. But I'm pretty sure there are papers published where people have done just that, so they must be much better at running gels than I am! I bet they had really nice gel-rigs.

Good luck!

- Sean

[nucacidhunter]

If you are preparing libraries from many samples which will leave you with multiple batches, I would suggest running all of them in the same gel to increase chance of selection of Rad-tags with the same length. This should improve tag overlap among samples. Manual gel size-selection also will result in broader size range than Pippin so you will need more sequencing for good coverage.

[ATϟGC]

Here is an article that may be of interest to you if you are looking to decrease costs and improve sequence quality:

https://academic.oup.com/mollus/article/82/1/104/2460141/Low-cost-ddRAD-method-of-SNP-discovery-and

https://www.researchgate.net/publication/282895172_Low-cost_ddRAD_method_of_SNP_discovery_and_genotyping_applied_to_the_periwinkle_Littorina_saxatilis

Essentially they use adapters on each end with the Nextera sequence so that dual indexing can be accomplished with PCR, which enables 384-plexing with the illumina indexing set.

This is essentially the same approach as many amplicon sequencing protocols and many lab order their own nextera-tailed primers/adapters and dual-indexing primers to save even more on costs.

Kess et al also use a custom primers during Illumina sequencing that spans the common restriction sites of both enzymes, which pretty much eliminates low-diversity problems.

[SNPsaurus]

I think RAD-PE is pretty inefficient as it takes skill to make the reads evenly distributed across the locus length (I say this as an author on the paper... not trying to be mean!) so there tends to be spikes in coverage. You most likely do not need a lot of sequence at each locus anyway. The first SNP at a locus is great... it's a new marker! The second has little extra value and many analyses ask that linked SNPs be thinned and only a single SNP at each locus be kept. RAD-PE, with it's 500 bp of sequence at each locus, or PE ddRAD with 200-300 bp might be overkill depending on the genetic diversity of your populations.

The precision of the size selection for ddRAD is especially important if you are trying to reduce the number of loci and a small size range is used. If the size range is 100 bp and the size accuracy is +-10bp, then that is a significant fraction of loci that will be lost between libraries made with different selection steps.

ATGC, what do you think your reference link means by "The first developed RAD sequencing methods used one enzyme to digest DNA and then a separate shearing process DNA to generate fragments of varying length, but sharing a restriction-enzyme cut site (Baird et al., 2008). Double-digest restriction-associated DNA sequencing (ddRAD) instead uses two different restriction enzymes to generate library fragments of genomic regions lying between the cut sites of these enzymes, further increasing the chance of sampling homologous genomic regions between sequenced individuals (Peterson et al., 2012)."

How does having two cut sites increase the chance of sampling homologous genomic regions? ddRAD has more nucleotides that can be disrupted by SNPs, if a frequent cutter is used then SNPs will create the frequent sites in the fragment and drop it from the library, and individuals in different libraries will have loci lost from having different size ranges selected. These all mean a lower chance of recovering a given loci between two samples, not an increased chance.

[nucacidhunter]

Thanks SNPsaurus for your comments. The paper also claims that their method is more cost effective. In my experience the largest portion of ddRAD expense is PCR reagents and labour and not the oligos. It seems that both of these factors are substantially increased in the this method.

[SNPsaurus]

I enjoy all the RAD variants and they are interesting how they solve different problems. In their case, they were sequencing on a MiSeq with a 600 cycle run, so the cost per read is going to be extremely high. The repetitive element in their genome made it important to try out different restriction enzyme combinations, and they didn't want to order a bunch of adapters for each try. So these selective pressures led to evolution of the method and a local optima for their situation!

Sometimes for publication, authors compare their method to others in order to demonstrate why it should be published, and it is difficult to do that impartially. They need to show it is not a local optima but generally optimal to get the most interest. But this is difficult, since others will say, "why work hard to multiplex on a MiSeq, costing $1800 for 25MM reads, when it is cheaper to just run it on a HiSeq4000, costing $1100 for 400MM reads?"

[ATϟGC]

Hi SNPsaurus,

I think that they are just referring to the typical single-digest RAD method where restriction is followed by sonication or another shearing method to generate fragment size compatible for your sequencing platform. I also think they meant to say "a separate DNA shearing process" rather than "a separate shearing process DNA".

In terms of increasing you chances of sampling homologous genomic regions: sdRAD relies on random fragmentation so recovery of homologous regions is more dependant on your read count. Since ddRAD uses two restriction sites as well as size selection it results in fewer but more consistent sampling of potentially homologous regions (and therefore higher read depths per locus). In my limited experience with sdRAD and ddRAD I saw evidence for this as ddRAD gave us higher per locus read depths with lower sequencing effort. You are right though that polymorphisms at restriction sites will cause allelic drop out in some loci. However, I think that low per-locus read coverage with sdRAD due to random shearing is more of a problem than using two restriction enzymes.

[SNPsaurus]

I see what you mean. I guess I would describe that as "ddRAD can generate fewer loci than sdRAD, so more sequencing may be needed to get sufficient read depth with sdRAD". My biased view, though, is that sequencing amount can be planned appropriately, while the SNP-based loss of data with ddRAD does not have such a simple remedy. It isn't a problem when genotyping a highly related population, but diverse samples will be biased to regions of the genome lacking genetic variation.

[tugecko]

Thank you everyone! This was all really helpful, and I appreciate your detailed replies.