MiSeq (and 2500) owners better read this and beware

Update 25th April: Illumina have released a support bulletin "Best Practices for High Sensitivity Applications: Minimizing Sample Carryover", this is accessible through Illumina's website, or SEQanswers. You'll need an Illumina login!

Illumina say that sample carryover is more likely to have an effect on very low detection threshold applications. Internal testing found sample carryover is typically below 0.1%, representing 1 read in 1000. Keep doing your maintenance washes (water and Tween) and do not use other wash additives, such as bleach, Triton or other decontamination solutions. 

The post as it was...
 A rather hot topic popped up on SEQanswers a few days ago. A new member Harlon, posted about seeing library contamination and carry-over between MiSeq runs; i.e. you get reads from the last run in the current one! So far the post has had almost 700 views and 13 replies. At least two are confirmations that this is being seen in other labs.

Are barcodes broken?

We are going to check our MiSeq runs!

Fingers crossed we don't see this issue on our machine!!!

So what's the problem: How big this issue is going to be is difficult to see right now but I'd expect it to be either a flash in the pan or a major headache for Illumina. If the issue is widespread then all the people that bought MiSeq for clinical work are going to need a much more robust cleaning protocol between runs. The current washes already add an hour to the run time and the use of NaOH is likely to be detrimental to the fluidics of the instrument.

Unfortunately many labs are using MiSeq to detect low prevalence mutations in cancer samples, at or even lower than 1%, and we're hoping to push that lower and lower. However the 0.2% carryover reported by Harlon would suggest a sensitivity of detection limit of at least 1% and possibly more like 2% if we are conservative in not mis-reporting contamination as clinically relevant mutations.

Imagine a patient with a BRAFV600E mutation that clearly requires Vemurafinib treatment. Another patient is sequenced on the next MiSeq run and appears to be BRAF normal, but with a 1% mutant BRAF, what does that result mean for treatment? We're still very much in the early days of NGS for clinical decision making. But we can look at parallels with forensics and STR-profiling. A 1% contamination rate would simply not be tolerated, people might actually to to jail for running such sloppy forensic services.

The future is suddenly not quite so bright.

What can you do about your data: We are thinking about the best way to monitor carry-over. We use a demultiplexing script that can identify all barcode sequences present in the index read rather than simply demultiplexing against a manifest. I had asked a while ago what all the non-demultiplexed reads were, even though a good run only has a few percent. It seems it's not all sequencing errors after all. But we did not think to look back at previous runs and see if the contaminating indexes were coming from previous samples.

I hope we can quickly introduce a check that compares index reads form run to run and looks specifically at how much carry over is happening. once we have a good idea of how bad the issue is then we can think about the best way to resolve it. I'm sure Illumina' engineers will be burning the midnight oil for the next couple of weeks coming up with a fix.

What about 2500: The same issue is likely to plague 2500 rapid run users as well, as the same on-board clustering is used. In hindsight using a fixed tip for introduction of the sample to MiSeq and HiSeq 2500 was not the optimal choice, but I don't see an easy way for Illumina to make this a disposable option. And I can't see an easy way to guarantee elimination of carry over without using a disposable tip.

Even a disposable tip may not be enough if residual sample is left in the fluidics lines on the MiSeq and HiSeq 2500. I'd expect this to be vanishingly small, but someone is going to need to design some careful experiments to prove just how much of an issue this is.

PS: please don't comment here but add to the SEQanswers post instead.

PPS: one fix could be to use many more indexes than we do currently. Anyone using PCR-added indexes like Nextera or TSCA uses a pair of i5 &i7 indexes that together give a unique sample barcode. Illumina use 20 oligos (8xi5 & 12xi7) to get 96 indexes, increasing this to 40 allows us to get 384 indexes, 80 get to 1536 and 2560 index oligos available as sets of 96 indexes would allow almost 100,000 unique barcodes. That should be plenty for most labs to only use each one once per year!