MiSeq 16S amplicon bad quality

[fish]

Hi,

I ran a library on the Illumina MiSeq at our institute with the V2 Kit (2 x 250 bp) and got rather bad read quality. The library contained 16S rRNA amplicons, the v4 region, of 79 fish skin samples. We did the dual indexing strategy described by Kozich et al. 2013.

The average quality is not too bad (73.9 % >=Q30) but the heatmaps looks terrible. I attached you the detailed information.

Normally I use USEARCH for merging the pairs according to quality check and when trying this I get empty output files and it tells me all my reverse reads are too short. Even when changing the parameters to a minimum of strictness the reverse read goes to the trash.
After usearch I would have followed the mothur pipeline. Interestingly, mothur merges my pairs since their algorithm is not very strict. However, I don't trust my results at all.

I'm in contact with some MiSeq experts and they rather assume the library itsself to be the problem (or the DNA) than the seqeuncing run.
Since I had a lot of trouble with getting nice PCR bands it could be the library (but what can have caused problems?) and I have to deal with the bad output.
Now, I'm thinking of using only the forward read for the analysis but I might loose a couple of nts at the end.
Did anybody of you had some similar sequencing output and give me some hints how I could improve the quality? Maybe I have to increase the diversity? We spiked in 10% phiX.

Thanks a lot!

[GenoMax]

It is not great data but I have certainly seen worse.

Have you tried to use a different merging program (bbmerge, FLASH instead of USEARCH)?

[fish]

No, I haven't.
I would have liked to stay with usearch since I processed my other libraries with that and I can then better compare the results. However, it seems not to work so I'm open for any suggestions.
I will have a closer look to the ones you mentioned.

[GenoMax]

If other tools also do not work then you may have a bad set of libraries.

[microgirl123]

What really strikes me as odd is that your cluster density looks decent, but your clusters passing filter is only ~50%. How big are your amplicons? Were there larger non-specific bands present in your PCR?

[fish]

Hi,

my amplicons were about 255 bp long and yes, I had some larger unspecific bands in my PCR. However, I ran the product on a gel, cut the expected band and did a gel purification to get rid of them.

Hope that helped.

[bunce]

I would have a look at the thumbnail images. Are they clean? - for every run we collect the first 4 images as this is sometimes key to trouble shooting. I would be suspicious of the clustering density - while 872K 'looks' like it is in the 'sweetspot' this number can in fact be erroneous. Compare the thumbnail images with a similar (density) and see if they look similar.... As genomax and microgirl point to there could be an underlying library issue at play here but IMO the low %PF is more indicative of a clustering issue (or perhaps a chemistry issue?). There look to be some short amplicons in the mix here...(based on the Q-score heat maps) - and when this happens libraries are difficult to quant making it easy to over cluster. Anyway thats my 2cents worth! - cheers Mike

p.s. - The PhiX error rate looks normal - but it is not present in read 4?.

[Brian Bushnell]

With libraries that have quality problems and long overlaps, I suggest using the "vloose" flag for BBMerge:

bbmerge.sh in1=r1.fq in2=r2.fq out=merged.fq outu=unmerged.fq vloose

I didn't know that USEARCH had an overlap-merging tool, but I'm finishing up a paper comparing BBMerge to all other mergers, so I'll add that in. I've never seen USEARCH evaluated or compared anywhere.

[microgirl123]

That is really strange that nothing aligns with phiX in Read 4 (your reverse read). It looks like something went really wrong there.

[fish]

Thanks for your help, so far!

The absence of phiX in read 4 is strange, indeed, and we're in contact with Illumina. I'll keep you updated.

With Flash I was able to merge up to 90% of the reads which looks, based on the bad reverse read, not really convincing to me...

[bilyl]

I haven't read the method from the paper, but did you use custom primers? It is a common mistake to use the wrong primer blend for read 2. PhiX uses the "PE Read 2" primer, which is different from the "Multiplexing Read 2" oligo. Because the Read 1 is common, then you will see PhiX in read 1 but not read 2 if you blended your primers incorrectly.