Bioinformatic analysis
The sequence reads were bioinformatically processed on the Genobioinfo computing cluster (GenoToul, Toulouse, France). Forward and reverse reads were assembled using the program illuminapairend from theOBITools v1.2.11 package (Boyer et al., 2016) and the resulting sequences were trimmed for primers using the program cutadapt (Martin, 2011). Strictly identical sequences were clustered together using obiuniq (dereplication step). Amplicon Sequence Variants (ASVs hereafter) occurring fewer than 100 times across all samples were discarded. Additionally, sequence length filtering was applied specifically for each primer set: 40-100 bp for Teleo ; 140-200 bp for MiFish ; and >100 bp for MG2 . To denoise the dataset, we used the obiclean function,keeping only sequences that were more frequently “head” or “singleton” than “internal” in the global dataset, with “internal” reads being potential PCR substitutions or indel errors (Giguet-Covex et al., 2014). The MetaBar package v1.0.0 (Zinger et al., 2021) was used to lower tag jumping and remove contaminants through detection of ASVs whose relative abundance was highest in negative controls. Finally, to remove false-positive results, we discarded not-shared occurrences among technical replicates (Robasky et al., 2014) (i.e., detected in only one of the three PCR replicates).