Terrible MiSeq PE R2 results !

[JackieBadger]

Hi,

We recently upgraded to the latest MiSeq software and hardware.
Read 1 of our 150bp paired end ddRAD data set came our fine.
However read 2 has very bad quality and very high GC content.

In the run previous to the upgrade we saw poor qualities in read 2 also, but all were usable.
This latest runs r2 is garbage.

Any clues to why?

Cheers

[Hanne]

Hi!

I'm sorry not being able to help you with this problem... I hope someone else can, good luck! :-)

We are also trying out this protocol in our lab. We will also use the MiSeq, but haven't sequenced yet. Can I please ask you some questions regarding the library prep?

Cheers,
Hanne

[mcnelson.phd]

I haven't dealt with RAD-seq on the MiSeq at all yet, so this is just my best guess, but what is possibly happening to your runs is the same error that's affecting all limited-diversity sequencing libraries on the MiSeq post-upgrade.

The problem is due to an upgrade to the RTA software and a change in how phasing/pre-phasing values affect overall base quality. In libraries where the base distribution gets really skewed during the first few cycles, if RTA estimates that phasing/pre-phasing reaches 0.4 or greater then from that point on it assigns really poor quality scores to each base call for all subsequent cycles. The way to get around that is to use a higher spike of phiX to even out the base distribution more, or to use a hardcoded set of RTA parameters for phasing/pre-phasing and the crosstalk matrix.

There is some debate over whether or not the actual basecall is good or not, aside from the given qscore, but that's another issue. Overall Read 2 will always have lower avg. quality because FWHM will increase as a result of the second strand re-synthesis. We've been doing our PE 500 cycle genome runs as 261/241 instead of 2x251 because the last few bases of Read 2 are always ~Q10 while the last bases for Read 1 are usually at least Q15.

[jesweber711]

Hanne brought this potential issue to my attention a few weeks back, but I haven't had a chance to look into the MiSeq upgrade. That said, if the upgrade indeed deals with base composition at the start of Read 2 runs, then the published ddRAD protocol won't be suitable for paired-end reads without doing some fiddling. The invariant restriction site will be the first sequence of ddRAD Read 2 data.

I'm not an Illumina guru, but I can see 2 immediate ways around this:
1. Stick to single-end reads and include at least a few different barcodes in each MiSeq run to even out the base composition.
2. It wouldn't be difficult to re-design the P2 adapters so that the index is part of the regular Read 2 sequence (the same way ddRAD incorporates the Read 1 index before the P1-associated restriction site), just move it between the Illumina primer sequence and the restriction site. We avoided doing this in the initial protocol because it was more cost effective to add at least one of the indexes during the PCR step. If you don't mind ordering 2 sets of costly adapters (48 X P1) AND (12 X P2), you can add all the indexes during the ligations. Everything else should still work and the base composition will be fixed by having at least a few different P2 indexes in each run.

Overall, I think the best fix would be to override the Read-2 base composition quality control in the MiSeq software, probably by hardcoding the expected restriction site bases. I'd be interested in hearing how this could be accomplished. Increasing the phiX dose would also help, but I'd be hesitant to throw away sequence unless it's absolutely necessary. Anyway, I'll think about this more and will ask around about possible software fixes.

Cheers-
Jesse

[Hanne]

Hi all!

I have delivered one sample (8 samples indexed) to sequencing (MiSeq 250 PE, lib mean 461 bp) and expect the data from this run in a few weeks. I have asked everywhere for help regarding the problem due to the monocromatic motif, thanks again Jesse! Illumina Illumina Technical Support recommends to apply hard-coding, “Altering the MiSeqConfiguration”. This means taking matrix, phasing and pre-phasing values from a good PhiX run and using these for running low diversity samples. They gave some instructions on how to do this, and we will try this out. Illumina awere us that this is not a supported Illumina workflow, so may require some optimization and comes with no guarantees. With hard-coding applied you may be able to just run the samples at a low cluster density (500-600K/mm2) and not spike-in so much PhiX (it may even work with no PhiX), which will help with the yield. We will go for 10% PhiX and I can update you guys, if you are interessed?

Hanne

[JackieBadger]

Yup we figured out the issue and it was indeed due to the lower complexity of the first 12 bases of read two from the ddRAD library. We spiked in ~50%PhiX and got much better results. Future libraries will include greater complexity in R2 to mitigate the issue.
Thanks for the input.

[GenoMax]

We spiked in ~50%PhiX and got much better results.
Thanks for the input.

Has that allowed the quality scores to recover to a normal level (inclusion of PhiX)?