Evaluating bioinformatics pipelines for population‐level inference using environmental DNA

Abstract Environmental DNA is mainly not only used at the interspecific level, to quantify species diversity in ecosystems, but can also be used to quantify intraspecific genetic variability, thus avoiding the need to sample individual tissue. However, errors in the amplification and sequencing of eDNA samples can blur this intraspecific signal and strongly over‐estimate genetic diversity. Existing bioinformatics pipelines therefore need to be tested to evaluate whether reliable levels of intraspecific genetic variability can be derived from eDNA samples. Here, we compare the ability of twelve metabarcoding pipelines to detect intraspecific genetic variability combining five programs. All pipelines have common pre‐processing steps, a processing data step using programs among obiclean; DADA2; SWARM; and LULU. An additional chimera removal step is also investigated based on two programs (VSEARCH or DADA2). The case study was the natural intraspecific variation within Mullus surmuletus in experimental settings. We developed specific primers for this species, located on the mitochondrial D‐loop fragment (barcode MS‐DL06). Thirty‐nine individuals were collected from the Mediterranean Sea, placed into four aquariums, and their DNA was sequenced on this marker to build an intraspecific reference database. After filtering the aquarium water, DNA was extracted, amplified, and sequenced using the primer pair developed. We then quantified the number of true haplotypes returned by each pipeline and its capacity to eliminate most of the erroneous sequences. We show that the program DADA2 with a two‐parent chimeric sequence removal step is the best tool to estimate intraspecific diversity from eDNA. Furthermore, our approach was also able to detect true M. surmuletus haplotypes in two eDNA samples collected in the Mediterranean Sea. We conclude that the combination of an appropriate intrapopulation barcode and a denoising pipeline like DADA2 with a chimeric sequence removal step is promising to make population‐level inference using environmental DNA possible.