Normally, when I work with pair-end 16S Illumina libraries covering a single hypervariable region (200 to 300 bases long), I simply trim low-quality tails with trimmomatic, join forward and reverse reads with fastq-join/vsearch and run closed-reference OTU-picking vs a database of full-size 16S sequences. This time I've come across amplicon libraries covering two consecutive hypervariable regions spanning around 500 bases in total. Given that the libraries were sequenced on MiSeq with 300-long reads and that the first 30 of them are actually adapters and overhangs, the actual sequence shrinks to around 270. Furthermore, the last 40-50 bases are poorly read and are usually removed during even the mildest trimming. Therefore there is no way I can get a large enough (if any) overlap between forward and reverse reads to join them. I've searched the web on how to run closed-reference OTU-picking with unmerged reads to no avail. Can you recommend something better than ditching one set of reads altogether? I wouldn't want to loose that information.