Hello,

I am no expert in the bioinformatic process but this paper (a bit old now) can still give you some insights / answers on methods that might have been enhanced now.

Also, with which sequencing machine did you sequence your samples ? I attached a technical paper comparing the old 4 - channel and the 2 - channel regarding base-call performances.

I hope it will give you some answers, good luck with your dataset.

What exactly do you mean by "crosstalk?" Which platform are you using?

@ Mark - by crosstalk I mean reads for OTUs, which should be exclusive for one sample on a MiSeq run, but we find them also in other samples. We are experimenting with USEARCH and VSEARCH pipelines.

Hi Jerome,

Three points that come to my mind:

1) What is the number of different bp in the tags used to label those samples? Could be simply tag-jumps which is known well. Do you apply a frequency filtering threshold?

2) Do you find the unexpected OTUs in the replicates (important to have replicates for that)?

3) Can you exclude real contamination in the lab? Maybe a molecule physically jumped in...

best Florian

Hi Florian,

1) we use the standard Illumina i5/i7 tags (e.g. TCGCCTTA or GCGTAAGA) - I will check for the frequency filtering.

2) We did not test replicates so far - but that might be a good starting point - Do you mean to test which index combination could be the reason for the "crosstalk" by apllying replicates? E.G. the same sample indexed with the different indices?

3) Contamination should not be the reason - (we use fresh aliquots and filter tips for all steps)

I will give it a shot though!

All the best - Jerome

Hi Jerome,

For ways to reduce "crosstalking" I suggest reading the papers attached.

Basically crosstalking can be reduced by using matching forward and reverse tags (Schnell et al.) or you could use a single PCR amplification step to incorporate all necessary technical sequences into your amplicons (i.e. sequencing adaptors, sequencing primers and tags) (Murray et al.). The later avoids the use of multiple PCR steps in the library preparation which will also reduce the amount of chimeric sequences.

Hope it helps.

Have you seen R Edgar's paper on the biorxiv? Have not seen it published peer reviewed yet, but might be worth looking into.

https://www.biorxiv.org/content/early/2016/11/19/088666

This is a huge problem for eDNA studies and we don't openly discuss this issue much, rather we remove data until we think we have removed the effect. I think it comes from the dark cycles of bridge amplification on the flow cell. I have a hunch that the longer the amplicon the more of a problem this is and is especially seen in non optimal clustering conditions. There is only so much space on the flow cell and the resolution of the imaging is only so good. Look at our paper disscussing phiX contamination.

http://onlinelibrary.wiley.com/doi/10.1002/ece3.2186/full. PhiX has no indexes but gets randomly assigned to samples. It's added at the last stage of library prep so no other possibility for contamination.

There are no real solutions yet that I know of. I think there should be some kind of isolation by distance effect that can be used based on the coordinates of the clusters, but I have yet to find time to test this. Save your flow cell images if you can!

Kristy

Hi Jerome!

Send me an email and I will do my best to help you out with your jumps and/or bleed: [email protected]. Have a great weekend!

Kristine

Jerome, can you be certain that the "contaminating" OTUs should truly be exclusive to one sample? In other words, is this some kind of reconstruction experiment where you spike a control into only one sample? If you are getting true "cross-talk", you should have another read with the same barcode produced from a nearby coordinate on the flow cell. Have you checked so see if this is the case? If you do, indeed, have bleed-over from nearby polonies, maybe you can filter the reads to remove ones having the same barcode that map too close on the flow cell. With hundreds of samples loaded on a flowcell, this should not occur very often...

Thanks for all the answers. We are performing some tests with MID tag equipped forward and reverse primers in order to strengthen the power of "over all indexing". I guess this will help eliminating unwanted OTUs from other samples - @Mark - yes that did happen in some cases - here OTUs had reads in every single sample - like a smear on gel ranging from the left to the right side.