Failed Illumina lane

[JoanaIEE]

Hi everyone
Our group has recently sequenced 8 RADtag libraries on Illumina HiSeq 2500.
7 of the lanes were very good but one of them was a complete failure. Even though the cluster density was good (~650k/mm2), most of them failed the filter. The 47 million reads we recieved were of very bad quality: 60% of them had an uncalled base (N) at the 7th position and after ~40 bp the quality drops below Phred score 10.
The library had a good concentration and fragment length distribution.
We have no clue what may have gone wrong and we would be very happy for any hint. I have attached some more information, FastQC and SAV plots.
Thanks a lot in advance for any help!
Best,
Joana

[kmcarr]

I would suspect a fluidics problem (e.g. clog or bubble) in lane 2 during cycle 7.

[SNPsaurus]

Just so I understand, the run wasn't done in rapid mode, but the traditional mode with 8 lanes? And it didn't have a control lane with phiX, but rather phiX was spiked into each lane? Were different libraries loaded into each lane?

It very well could have been a fluidic problem, but the fact that the problem first appeared at the invariant nucleotide of the restriction site makes complexity of the read another possibility. Or a combination of the two. I was surprised that the other lanes with cluster densities approaching 1000 did as well as they did. We (in my academic side) found it worthwhile to aim a bit lower than suggested with RAD libraries. I'm also surprised using all 8 lanes without a control lane worked out as well as it did. Sorry to not have any real advice!

[JoanaIEE]

Hi SNPsaurus and kmcarr

Thanks a lot for your comments.

Just so I understand, the run wasn't done in rapid mode, but the traditional mode with 8 lanes? And it didn't have a control lane with phiX, but rather phiX was spiked into each lane? Were different libraries loaded into each lane?

Yes, it was a traditional mode run with 8 lanes. Each library containing several barcodes was loaded on one lane and spiked with 20% PhiX.

It very well could have been a fluidic problem, but the fact that the problem first appeared at the invariant nucleotide of the restriction site makes complexity of the read another possibility. Or a combination of the two. I was surprised that the other lanes with cluster densities approaching 1000 did as well as they did. We (in my academic side) found it worthwhile to aim a bit lower than suggested with RAD libraries. I'm also surprised using all 8 lanes without a control lane worked out as well as it did. Sorry to not have any real advice!

Yes, it does not seem like a coincidence that the problematic first nucleotide is an invariant one in the library. But all other libraries also had the same base at position 7 and it was no problem for the other libraries even though most of them had a higher cluster density. At another sequencing center they also successfully use 20% PhiX for RAD libraries, that's why this amount was chosen.

[TonyBrooks]

Filtering is done using data from the first 25 bases. One failed test is allowed for every cluster. If the same cluster fails a test across two different cycles, then the cluster fails filter. If you get BMS (bubble causing mis-focus across a swath) within the first 25 cycles this causes problems as every cluster will fail tests. It means you are inherently at risk of losing clusters to the filter. Can you post the thumbnails from the dodgy cycle (7).

[HGalvez]

Had a similar problem here and we attributed to a BMS failure in the first cycles. As other people mentioned in this thread, it seems there was a problem in cycle 7 for lane 2. You should be able to see what it was in the SAV. If you saved the thumbnail pictures the problem should be even more evident.

BMS are a major source of headaches for anyone working with the HiSeq platform and I can tell you from experience that the 2500 seems more likely to fail for some reason than the 2000. If you call illumina to tell them about your problem, they will tell you that you should change gaskets often, but it doesn't seem to be failsafe.