Erratic cluster numbers

[andibody]

Hi,

Since ~ 2 months we are getting erratic cluster numbers on our GAs (cBot) even if we check the libraries with RT-PCR. Before we were using 7pM (~300.000clusters /tile) and were getting reproducible, nice results. Now a 7pM sample might result in more than 900.000/mm2 clusters. Also single lanes might totally fail without obvious reasons (not even for techsupport).
I just saw a recent similar thread, but wanted to ask more specific if other/more people are experiencing the same problems.

Thanks
Andi

[GW_OK]

If you've updated your software that could be the issue. The new software uses a different cluster finding algorithm that's better at identifying clusters than the old one. Our FAS said we ought to re-titrate all our libraries, but since we don't titrate them anyway it's a bit of a moot point. We run ~10-10.5 pM with pretty good results.

For the lane failures, is it always the same lane?

[NextGenSeq]

If you spike in PhiX control at 1% or less, it will tell you if its the library or a technical amplification issue.

[Efortier]

For the lanes that totally failed for no apparent reasons, we had similar problems a few months ago. The problem was the cBot plates, more precisely the reagent HP5 (0.1N NaOH). For some reason during shipping on dry ice, the NaOH pH changed due to the high amount of CO2 produced by the dry ice. This resulted in bad primer hybridisation on the flowcell because the NaOH was not at pH above 12. To work around this problem we just made our own 0.1N NaOH and replaced the HP5 with our NaOH. The new cBot plates should have red foil sealing the HP5 instead of the silicon-type tubes cap. This should fix this issue I think. You can also recover a bad primer hybridisation by redoing the primer hybridisation on the cBot with the appropriate reagents.

Hope this helps.

[huhrik]

have you checked raw images for failed lanes?
how do they looks?

[andibody]

Indeed, it started happening after we upgraded the software. Could it be that the new image analysis resolves more clusters but at the same time is less unforgivable to high cluster density?
@NextGenSeq: if the software can't resolve the images it won't also be able to do so for PhiX clusters.
@huhrik: they are indeed denser than 'good' lanes but far from being 'fully covered'

[simonandrews]

By any chance are you using libraries which have very biased sequence composition at the start of the first read? This can cause problems with cluster detection which can look like you're getting low cluster numbers.

[andibody]

no, the density problems were observed even on whole genome samples.