A possible problem with multiplex tags just occurred to me

[kmcarr]

We just completed a run where an interesting artifact of GA sequencing surprised us a little. A researcher wanted to sequence a reduced representation of a genome. He digested genomic DNA with AluI, ran the digest out on a gel and isolated fragments in a particular size range. That DNA was prepared for sequencing without further fragmentation. This means that the 5' end of all the DNA strands started with 'CT'. The consequences of this did not occur to us until we saw the images of the first cycle of sequencing. The extreme bias in the base composition of the first two cycles completely flumoxed the cluster identification routines of the Illumina image analysis software. No problem though as I can rerun the analysis from scratch with the pipeline but only using cycles 3-36.

This got me thinking. Suppose you have designed your own multiplex adapter oligos, placing the bar code immediately downstream of the primer site. Depending on the base representation in the mixture of bar codes you are using this could lead to a problem like the one we encountered above, but you can't simply ignore those cycles since you will need those bases to sort the samples.

Have I made any sense at all? Has anyone encountered this situation using bar codes or have you developed protocols to make sure you avoid it?

[swbarnes2]

I thought that was the purpose of the control lane; that even if your other lanes had weird biases, that the software would trust that those lanes were correct if the control lane was showing the right ratio of letters.

[kmcarr]

Yes, the control lane can be used exclusively to calculate the crosstalk matrix and phasing numbers. But there was a significant problem with the cluster identification. The cluster template is determined in the first three cycles (with the new, v2.4 software, prior to that it was just the first two cycles) and I was told by the tech support person that the software needs to be seeing some signal in all four channels to identify cluster locations. That was my main concern about samples in which (nearly) all of the fragments have identical bases for the first cycle or two.

However I have now found that in addition to 5' ends of all the fragments being identical the flow cell was horribly overloaded, making it impossible for the software to isolate clusters even if the base composition was perfectly balanced. I guess we'll be rerunning the sample at a much lower concentration and I'll see then what effect the biased 5' ends have on cluster identification.