Dual-index barcode read failures

[james hadfield]

Hi All,
We've seen a lot of read 2 failures on our HiSeq and on externally supplied data. The quality of the second index is terrible and we lose most of the reads as they cannot be demultiplexed.
Having looked into this we are now confident the root-cause is acidification of NaOH during the run, and that this means the template removal step before index reads is compromised. When we check NaOH on long-read (PE100) runs the pH has dropped to less than 10.

Has anyone esle seen this? I'd lie to get an idea of how widely this problem is being seen as NaOH has been a fly in Illumina's ointment for ever! Such a simple reagent should not be the casuse of a £5000 run failing after a weeks seqeuncing on a £450,000 machine!!!
Thanks.
James.

[thomasblomquist]

This is one reason why we now have placed our own barcodes inline flanking our sequence of interest. They are read out in read 1 and 2. And based on these reads we demultiplex the sequences into sample specific bins. The index reads have not always worked well for us, as you are dealing with. And this can be particularly frustrating with good quality read 1 and 2 data.

-Tom

[GenoMax]

Index reads are particularly sensitive to sample overloading. One can get away with overloading on the actual reads but the index reads fail when samples go over the loading cliff.

[koadman]

Hi James,
Yes, I've fallen victim to this problem previously, see the post from February this year:

http://seqanswers.com/forums/showpost.php?p=96466&postcount=14

I've been working mostly with the miseq since then and haven't had these issues (though our miseq went through three episodes open-heart valve transplant surgery before stabilizing).

p.s. where's pmiguel! :-)

[pmiguel]

Hi James,
Yes, I've fallen victim to this problem previously, see the post from February this year:

http://seqanswers.com/forums/showpost.php?p=96466&postcount=14

I've been working mostly with the miseq since then and haven't had these issues (though our miseq went through three episodes open-heart valve transplant surgery before stabilizing).

p.s. where's pmiguel! :-)

I'm here, Koadman. Why do you ask?

--
Phillip

[NGS_curie]

Hi James,

the index 2 read just finished on our run with dual-indexed Nextera XT libraries and we see exactly the same bad quality you mentionned. Only 50% of the samples are correctly identified with 95% of index1 >Q30 and only ~50% of index2 >Q30.
In our case, we first made a mistake in loading the reagents for index 2 so we had to re-do the index 2 read. We put fresh NaOH but this does not seem to prevent from index2 failure reagrding the low quality we get.
This does not happen with Rapid runs as we did some runs with this configuration.

Does anyone know if there is a way to use index sequences with lower quality to try to rescue some read identifiation ? It seems that only Q30 sequences are kept for read identification and I believe regarding the low number of index 2 that even with one mistake in the index we could identify the right samples with high confidence.
I still do not understand why illumina does this dual indexing process whereas 96 unique index 1 would be much more easy and efficient like Agilent does.

Thomas

[pmiguel]

I still do not understand why illumina does this dual indexing process whereas 96 unique index 1 would be much more easy and efficient like Agilent does.

Thomas

I prefer dual indexing as it allow the processing of larger numbers of samples with fewer oligos -- for PCR-based methods for adding indexes.

Would be best if Illumina discontinued high output chemistry. Or retrofitted it to seamlessly deal with the dual bar codes like Rapid/MiSeq workflows do.

--
Phillip

[pmiguel]

Do you use on-board analysis or have access to off-line analysis tools? I assume this is a HiSeq run. I have been helping another user with off-line analysis for a similar issue but before I point you to that thread I wanted to check.

Sometimes it is just a single cycle that is low quality. It is fairly easy to direct the demultiplexing aspect of CASAVA to omit a particular cycle from consideration. But, yes, that is off-instrument.

I did not realize it was even possible to do on-instrument fastq file generation on the HiSeq.

--
Phillip

[GenoMax]

I did not realize it was even possible to do on-instrument fastq file generation on the HiSeq.

--
Phillip

Posting before the coffee kicks in is not a good idea. Thanks for the reminder.

[pmiguel]

Posting before the coffee kicks in is not a good idea. Thanks for the reminder.

This is only tangentially related to this thread: have you tried using BaseSpace for a HiSeq run?

In our environment, seemed superfluous. So I never even asked for the BaseSpace module to be installed. But seems like an interesting idea.

--
Phillip

[GenoMax]

This is only tangentially related to this thread: have you tried using BaseSpace for a HiSeq run?

No I have not. Will PM you.

In our environment, seemed superfluous. So I never even asked for the BaseSpace module to be installed. But seems like an interesting idea.

--
Phillip

AFAIK BaseSpace service is automatically installed on HiSeq's workstations with the new control software (and it may be active by default!). Have you looked in the windows services to see if it is running on your HiSeq?

[pmiguel]

No I have not. Will PM you.

I don't thing it is that far off-topic. Can discuss it here. Guess we could start another thread, if necessary.

AFAIK BaseSpace service is automatically installed on HiSeq's workstations with the new control software (and it may be active by default!). Have you looked in the windows services to see if it is running on your HiSeq?

No, it isn't running. Our engineer didn't install it. And I never bothered afterwards.

--
Phillip

[james hadfield]

I thought I should report on the problem we were seeing with read 2 index failures.
It turns out that we were seeing the affect of acidification of NaOH on the instrument, this meant the first sequence reads did not get stripped off properly and caused the drop in index read quality. We've taken to pausing all long-read runs before indexing and chaning NaOH.
The HiSeq 2500 is not affected in the same way.
It's a shame that such a simple thing (NaOH) is still killing Illumina sequencing. Who remebers the paired-end kit problems of 2009 or scan mix of 2007?