Bioinformatic treatment
The bioinformatic treatment of sequence data was performed using the OBITools software suite (Boyer et al., 2016). First, forward and reverse reads were assembled using the illuminapairedend program, keeping only sequences with an alignment score higher than 40. Aligned sequences were assigned to the corresponding PCR replicate using the programngsfilter , by allowing two and zero mismatches on primers and tags, respectively. After sequence dereplication using obiuniq , bad-quality sequences (i.e. containing “N”), sequences whose length fell outside the expected size interval (below 45 bp for Bact02, below 68 bp Fung02 and below 36 bp for Euka02) and singletons were filtered out. The obiclean program was run to detect potential PCR or sequencing errors with the -r option set at 0.5: in a PCR reaction, sequences are tagged as “heads” when they are at least twice as abundant as other related sequences differing by one base. Only the sequences tagged as “heads” in at least one PCR were kept.
Taxonomic assignment was conducted using the ecotag program based on a reference database constructed from EMBL (version 136) by running the ecoPCR program (Ficetola et al., 2010). More specifically,ecoPCR carried out an in silico PCR with the primer pair used for the experiment and allowing three mismatches per primer. The obtained reference databases were further curated by keeping only sequences assigned at the species, genus and family levels.
Further data filtering was performed in R version 3.6.1 (R Core Team, 2018) to remove spurious sequences that can bias ecological conclusions drawn from DNA metabarcoding data (Calderón‐Sanou et al., 2020). More specifically, we discarded from our dataset MOTUs with a best identity <85% (Fung02, Bact02) or <80% (Euka02), observed less than five times overall or in more than one extraction or PCR negative control (Zinger, Bonin, et al., 2019a). Furthermore, we removed all MOTUs that were detected in less than two PCR replicates of the same sample, as they often represent false positives (Ficetola et al., 2015).