2.4 Sequence analysis
Raw sequences were assembled using FLASH (v1.2.11) with default settings (Magoč and Salzberg, 2011). The assembled files were subsequently demultiplexed, quality filtered using the quantitative insight into microbial ecology (QIIME) pipeline v1.9.0 (Caporaso et al., 2010). Operational taxonomic units (OTUs) were clustered with 97% similarity cutoff using UPARSE algorithm (Edgar, 2013). Taxonomic assignment was performed by RDP Classifier (Wang et al., 2007) against the SILVA 132 NR database (Quast et al., 2012) at a bootstrap cutoff of 80%. OTUs that were affiliated with chloroplast, archaeal, and unclassified sequences (not affiliated with Bacteria) were removed from subsequent analysis. To ensure that rare bacteria were not the result of sequencing errors, OTUs that present in <10 samples and/or possessing <10 sequences were discarded. All further data analyses were based on the rarefied OTU table at 18799 reads per sample, which was created according to the minimum sequence number of 16S rRNA gene amplicon reads per sample.