Abstract1. Metabarcoding of environmental samples has many challenges and limitations that require carefully considered laboratory and analysis workflows to ensure reliable results. We explore how decisions regarding study design, laboratory set-up, and bioinformatic processing affect the final results, and provide guidelines for reliable study of environmental samples.2. We evaluate the performance of four primer sets targeting COI and 16S regions characterizing arthropod diversity in bat faecal samples, and investigate how metabarcoding results are affected by parameters including: (1) number of PCR replicates per sample, (2) sequencing depth, (3) PCR replicate processing strategy (i.e. either additively, by combining the sequences obtained from the PCR replicates, or restrictively, by only retaining sequences that occur in multiple PCR replicates for each sample), (4) minimum copy number for sequences to be retained, (5) chimera removal, and (6) similarity thresholds for Operational Taxonomic Unit (OTU) clustering. Lastly, we measure within-and between-taxa dissimilarities when using sequences from public databases to determine the most appropriate thresholds for OTU clustering and taxonomy assignment.3. Our results show that the use of multiple primer sets reduces taxonomic biases and increases taxonomic coverage. Taxonomic profiles resulting from each primer set are principally affected by how many PCR replicates are carried out per sample and how sequences are filtered across them, the sequence copy number threshold and the OTU clustering threshold. We also report considerable diversity differences between PCR replicates from each sample. Sequencing depth increases the dissimilarity between PCR replicates unless the bioinformatic strategies to remove allegedly artefactual sequences are adjusted according to the number of analysed sequences. Finally, we show that the appropriate identity thresholds for OTU clustering and taxonomy assignment differ between markers.4. Metabarcoding of complex environmental samples ideally requires (1) investigation of whether more than one primer sets targeting the same taxonomic group is needed to offset primer biases, (2) more than one PCR replicate per sample, (3) bioinformatic processing of sequences that balance diversity detection with removal of artefactual sequences, and (4) empirical selection of OTU clustering and taxonomy assignment thresholds tailored to each marker and the obtained taxa.

No abstract