[illuminaGA]

Hi Harlon

Could you give me some pointers about how to detect the carryover contamination? Software, pipeline etc. Thank you in advance.

Hi All

We appear to be suffering from carry over contamination in our MiSeq runs - i.e. if we sequence a DNA sample in one MiSeq run, we see about the same sample in the subsequent run.

Measured in terms of reads, we see about 0.2% contamination run-to-run - i.e. if we see 10,000 reads of a given amplicon/barcode in one run, we'll see ~20 reads in the following run, even if that amplicon/barcode pair was absent from the prep.

Important notes about our workflow:
- Barcodes are added by PCR (we are using our own library prep, not Nextera, etc).
- We perform a post-run wash and a maintenance wash after every run.

I am quite certain that this is carry-over within the MiSeq, and that it is actually carry-over, and not simply barcode contamination within the primers. On the first run of an amplicon, it only shows up with its assigned barcode. It is also detected in the subsequent run. I'm quite certain this is also not laboratory contamination.

After speaking with Illumina, this is certainly feasible - they are aware of this issue, although they do see less run-to-run contamination than we see. They suggested that we do 2 maintenance washes between runs, which seems like a lot, and that we don't pour bleach into it, which was certainly not my plan.

Has anyone else had similar issues, and more importantly, does anyone know of any solutions?

Thanks!
Harlon

[Mercaptan]

I did a little experiment comparing read carry-over with our old regular washes (three cycle 0.5% Tween 20 wash) and Illumina's bleach wash.

Bcl2Fastq parsing of the raw data, checking for indices from prior runs, shows that the bleach wash reduces the carry-over from previous run from 0.018% (~3000 reads) to 0.00011% (23 reads). This was just comparing two runs, but that's a big reduction. We'll be keeping an eye on this in the future.

We're now doing the bleach wash as a replacement for our post-run washes. Followed by a regular three-cycle maintenance wash because we're paranoid about bleach getting into the next run.

[james hadfield]

Hi, I just wrote a new post on my blog, http://core-genomics.blogspot.co.uk/...ng-bleach.html which describes the bleach protocol and how to get it.

[bunce]

Hi Kristen,
Just one follow-up note about primers - I have had this response a few times about cost..... but non-HPLC purified primers from IDT are very cheap..... by the time you have repeated a few runs due to contamination they are paid for (not to mention the time and headaches spent chasing contamination). Contra to what NGS companies tell you there is no need to HPLC purify primers - if you screen for exact adaptors/MIDs any 'error' at primer synthesis is discarded (and most Indexes require 3 mutations to morph it into another index). In our (sensitive) workflows re-using indexes is a false economy... cheers Mike

[KristenC]

Hi Tom and Mike,

Thank you for your swift reply.
Tom, I really like your set-up with the internal standard as a competitor, already when I first read your paper a while ago. But we were at that time already running experiments within a project (I started halfway in the project), for which I could not implement your strategy anymore. I am now analyzing all the data and need to deal with the contamination as it is now. I will however use the results to improve our set-up for the next project, including the recommendation to use an internal standard to overcome these issues.

Mike,
We are going to implement the bleach post run for our new project, but will keep screening for previously used adapters to see if it makes a difference. We did see a big difference when we started doing the standby wash procedure after every run. Levels drop off to a very low level after 2 runs (less than 0.1%), but do remain at that level after that, which is why I wanted to try and use that level as a sort of background.

As for the set-up, we have a number of rules in place to avoid contamination in the sample prep, but there's always room for improvement. I will check where our wash solution are made, maybe we can improve on that.
We once had a primer fail for one of our targets, but decided to run the library anyway, and the target had 0 counts, which I thought indicated that we weren't doing too bad contamination wise ... :-).
I'm afraid our budget wouldn't allow us to order new indexes every time. I'll see if we can get some new ones in though, 24 may not be enough...


Thanks for your help already!

[bunce]

Hi Kristen,
Working in the field of ancient (and low copy number) DNA often in mixed samples - we are paranoid about this kind of contamination. We primarily do amplicon sequencing and a few years back after picking up carry over in 454 (and PGM...now MiSeq) we decided not to reuse index combinations..... this means we have a hefty primer bill, but contamination is bioinformatically screened-out as each library is unique.

We now do our sequencing on the MiSeq and in addition to the unique indexes we also do the new bleach "post run" wash.... I have no idea how effective this wash is (there are figures from illumina if you believe them!) but it is very easy to implement and should be your first option.

We do a few other things to minimise contamination such as making NaOH up in our pre-PCR area. From your description of the contamination (where you don't re-use tags for two runs) it may well be that your environment/set-up that is contaminated and may not be the instrumentation.

To answer your question about threshold cut-offs; this is next to impossible to put a reliable number on - the nature of background contamination is sporadic - Tom Blomquist's suggestion is a good one if you need to quantify it.... but our solution was to side-step the issue through unique index combinations.

hope this helps?
Mike

[thomasblomquist]

Hi,

Again reviving the topic...

We are running a targeted RNA assay on our Miseq, and also have that issue with the carry-over of previously run libraries. We have two sets of about 40 targets and we alternate those two panels each run. We use 4 adapters per run for sample libraries and screen for the other 20 adapters to check for contamination. Adapters used for sample libraries are not used for sample libraries for at least the following two runs. After each run, a post-run wash is performed followed by a stand-by wash.

By screening for the adapters that were not used in the present run, we can see that despite all our measures, we keep picking up reads that align to our targets, beit not exceeding Illumina guidelines. The distribution of those contaminant reads along the targets is similar to what we see in the actual sample libraries (High expressors have more 'contaminant' reads than low expressors).

But it makes me wonder whether a count of let's say 5 is an actual detected count, or whether that may be due to contamination and as such should be assigned an 'undetected'. For expressors that are consistently low, there is practically no contamination present in the unused adapters, but for those that vary between 0 and 1000 counts, a count of 5 or even 20 is hard to interpret.

So I wanted to ask if anybody has experience with setting a threshold below which you consider a target to be undetected based on contamination profiles. E.g. would you use a theoretical average contamination per target per run based on the levels in the other-than-sample-library adapters?

Hi Kristen,

Some time back you stumbled across our paper on targeted RNA seq: http://seqanswers.com/forums/showthread.php?t=35379

The use of the competitive internal standard molecules actually competes out and will provide a relative measure of the contamination. For example, it can be hard to know in PCR amplified libraries if 5, 10, 20 or even 100 or 1000 reads of a template is true. However, if you have a standard competitor present in the library prep (say 100 molecules), and no native template present during the prep, and your reads are 5 native to 1000 standard for that run (the 5 being contaminant carry over from the tubing on the illumina, you can calculate the actual number of molecules:

5/1000 * 100 = less than 1 molecule (0.5) of native in the actual reaction ~~~ likely equals contaminant.

This has been our approach. The competitor standard gives us a reality check to the biases of NGS library prep as well as carry over contamination for the Illumina platforms.

Regards,

-Tom Blomquist

[KristenC]

Targeted RNA assay

Hi,

Again reviving the topic...

We are running a targeted RNA assay on our Miseq, and also have that issue with the carry-over of previously run libraries. We have two sets of about 40 targets and we alternate those two panels each run. We use 4 adapters per run for sample libraries and screen for the other 20 adapters to check for contamination. Adapters used for sample libraries are not used for sample libraries for at least the following two runs. After each run, a post-run wash is performed followed by a stand-by wash.

By screening for the adapters that were not used in the present run, we can see that despite all our measures, we keep picking up reads that align to our targets, beit not exceeding Illumina guidelines. The distribution of those contaminant reads along the targets is similar to what we see in the actual sample libraries (High expressors have more 'contaminant' reads than low expressors).

But it makes me wonder whether a count of let's say 5 is an actual detected count, or whether that may be due to contamination and as such should be assigned an 'undetected'. For expressors that are consistently low, there is practically no contamination present in the unused adapters, but for those that vary between 0 and 1000 counts, a count of 5 or even 20 is hard to interpret.

So I wanted to ask if anybody has experience with setting a threshold below which you consider a target to be undetected based on contamination profiles. E.g. would you use a theoretical average contamination per target per run based on the levels in the other-than-sample-library adapters?

[pmiguel]

Seems like contamination has to derive from the fluidics that deliver the denatured library to the flowcell. Standard procedure to change gaskets (and, of course, flow cells). During this process wiping down the area where the gasket sits could be added prior to placing the new gasket.

Also you could use the "check" (flow check) capability of the instrument software to run extra 1N NaOH/water flows through the sample input ports. Seems like this could be done pretty quickly then followed with a few water flows.

--
Phillip

[thomasblomquist]

Reviving this thread in hopes that other users out there have found a way to solve this problem on the HiSeq2500 without rotating barcodes or clustering on the cBot. We have not, despite operationally switching to 2 water washes for every run.

I have barcode contamination data on literally hundreds of runs on many instruments where we run both sides around the clock. The problem is absolutely due to carryover from the previous run based on the barcode contamination patterns.

Any experimental wash procedures that are working better out there? Unfortunately, we are very sensitive to this problem, even in the 0.2-0.5% carryover range.

Any update on this? We will be ramping up our sequencing efforts and need to have this issue resolved as well. If not, we may jump back to an Ion Torrent approach. Even though it is more expensive, it does not have this very serious flaw.

-Tom

[nhunkapiller]

Reviving this thread in hopes that other users out there have found a way to solve this problem on the HiSeq2500 without rotating barcodes or clustering on the cBot. We have not, despite operationally switching to 2 water washes for every run.

I have barcode contamination data on literally hundreds of runs on many instruments where we run both sides around the clock. The problem is absolutely due to carryover from the previous run based on the barcode contamination patterns.

Any experimental wash procedures that are working better out there? Unfortunately, we are very sensitive to this problem, even in the 0.2-0.5% carryover range.

[MrGuy]

Even getting a single read of something errant is enough to shut. down. everything. Thus having even the slightest risk of detectable contamination means having proper controls in place to properly limit these contamination events as a known entity.

[thomasblomquist]

Some types of genetic variation are easily detectable beyond the typical single nucleotide variant limit of detection in sequencing. Translocations, indels, repeats, etc. Importantly, their relative abundance, even at very low frequencies can be clinically relevant.

[snetmcom]

I am just trying to understand this issue more. If the contamination is less than 1%, it seems largely irrelevant. Based on my experience, trying to call variants of that frequency will yield a huge amount of false positives. Isn't the sequencing error rate a much bigger problem that makes this a moot point?

1% contamination means you will have 150,000 reads of contamination in a normal run. Not 1% variant frequency.

It's not trivial for certain projects.

[id0]

I am just trying to understand this issue more. If the contamination is less than 1%, it seems largely irrelevant. Based on my experience, trying to call variants of that frequency will yield a huge amount of false positives. Isn't the sequencing error rate a much bigger problem that makes this a moot point?