MiSeq v2 vs. v3 kits - run variation

[fish]

Hi,

last year we sequenced two libraries of fish skin samples for one study, the V4 region of the 16S.
The first library was sequenced with the v3 kit because it was new and our sequencing service thought it's worth a try. However, for V4 I actually don't need the v3 kit since the amplicon is ~ 250 bp long. I received ok sequences but later had problems with an increase of unique sequences in the analysis pipeline, a well known problem very nicely described by Pat Schloss: http://blog.mothur.org/2014/09/11/Wh...nce-matrix%3F/.
So we sequenced the second library with the v2 kit and I included 15 samples from library 1 in there, to be able to correct for run variation. In fact, the 15 technical replicates were very similar in later analysis but when I checked the whole data set of 250 samples, I detected in part a large variation in the data according to the different runs. For instance, when I analyzed alpha-diversity, the factor run had the highest impact on the results, higher than the factors I was interested in.

Did somebody of you ever had the same problem?
I'm actually afraid that I can not publish my results because of that issue.

Thanks!

[thermophile]

Were different parts of the experiment run on both or were controls all run first then treatment later? If you compare the 15 repeated samples with something like partial-mantel, how similar are they?

[fish]

Hi,

I tried to distribute the respective groups of interest I wanted to compare evenly between both runs. The 15 replicates were run with the other samples, I had two runs in total. Did that answer your question?

I'm not sure what how you would apply with the mantel test, because I would need a second dissimilarity matrix to calculate the correlation, besides Bray Curtis distance or so of the OTUs of the 15 replicates. Maybe you can give me a hint.

[thermophile]

you'd cluster all of your samples into a giant OTU table, calculate your dissimilarity matrices not for each sample individually but on each treatment type on each run. Then do partial mantel comparing all the treatments on run1 to all the treatments on run2. you're hoping for non-significance