MiSeq amplicon: corrected int spike

[sharchaea]

My last MiSeq 16S amplicon run (2x250) and a rerun with a lower cluster density both showed large spikes in corrected intensity (>1000) during the second sequencing read cycles (see attached). This resulted in poor read quality (see attached). The lower cluster density for the re-run produced one big spike in intensity (>2000) at cycle ~410 (spike also present in original run). We used custom 16S rRNA primers that closely reflect the Kozich et al., 2013 dual-indexed primers (our forward read has a couple more degeneracies). This is our first time using these primers on a MiSeq run. We multiplexed ~90 samples, and the rerun used the same library prep, but just a different cluster density.

Here's some stats on each MiSeq run:

Original run:
* cluster density = 963
* cluster PF = 49.38
* %>=Q30 = 52.53 (read1), 56.57 (read2)
* % aligned = 8.9 (read1), 6.69 (read2)
re-run:
* cluster density=752
* cluster PF=52.23
* %>=Q30 = 61.68 (read1), 67.92 (read2)
* % aligned = 29.8 (read1), 28.63 (read2)
Can anyone help me troubleshoot what may be causing these large spikes in corrected intensity (which I hope will improve read quality)?

Thanks.

[acockburn]

I use the Kozich et al. primer set. I have never seen anything like this.

Could you give more info on the library generation?

Lately the reagents for the MiSeq has been crap and my Q30 have been terrible in the last 100 cycles (amplicon and genome libraries), but it seems like your Q30's are pretty low from the get-go.

[sharchaea]

Our library prep method follows Kozich et al.,:
* PCR with dual-indexed primers
* Pooling of replicate PCR reactions
* SequalPrep
* Speed-vac to concentrate the pooled samples for a gel excision
* Gel excision
* (optional) speed-vac to increase the concentration for library submission.

I've contacted Illumina tech support, and they think it might be a result of an issue with the MiSeq hardware. Apparently, a spike in corrected intensity can be caused by a transient bubble, a transient focusing error, or debris.