loosing reads after dada2 filteration

[colinbrislawn (Colin J Brislawn)]

Here is one last clue that may be related to your problem:

It's by benjjneb, who is the developer of DADA2, and related to the V3-V4 primers, like you are using:

@benjjneb

Yes, there is variation in lengths of 16S segments. It isn't large, but it exists. In particular, there are two modes of V3V4 length in nature, one at 460 nts and another ~440 nts (using these primers). So you'll want to make sure even the longer natural amplicons will sufficiently overlap after truncation.

If these 8 samples happen to have lots of microbes with the longer ~460 bp amplicon, then they need that extra length to merge.