Comparing different types of OTUs clustering methods
To analyze detection limits and taxonomic resolution of the three
methods used to infer OTUs, we plotted the Fungi-only tree along with a
heatmap of the average relative read abundance across all samples in
separate columns for OTU_A, OTU_C and OTU_S (Supplementary datafile
2). To explore the phylogenetic resolution of the three methods, the
ITS2 regions extracted from each sequence by LSUx (as described above)
were clustered using the same methodology outlined in Kõljalg et
al. (2013) to generate UNITE species hypotheses (SH) at 97 and 99%
sequence similarity: sequences were first pre-clustered at 80% sequence
similarity by VSEARCH, and then the sequences within each pre-cluster
were clustered at 97% and 99% similarity by BLASTCLUST (version
2.2.26; Altschul et al., 1990; Dondoshansky & Wolf, 2000). In addition
to these respectively lax and more stringent species-level thresholds,
we also generated approximately genus-level clusters using a 90%
(GH_90) similarity threshold (Tedersoo et al., 2014). We then mapped
the three ITS2 clustering levels onto the phylogenetic tree in order to
determine how many clusters were monophyletic and how well the three
different OTU generation methods captured diversity at different
taxonomic levels. Further, we estimated the abundance necessary for a
taxon to be detected as an OTU_A. For this we used the average read
abundance of OTU_S and OTU_C sequences assigned to GH_90, SH_97 and
SH_99 ITS2 clusters to identify the detection limit of DADA2.