The current capacity of environmental DNA (eDNA) to provide accurate insights into the biodiversity of megadiverse regions (e.g., the Neotropics) requires further evaluation to ensure its reliability for long-term monitoring. In this study, we first evaluated the taxonomic resolution capabilities of a short fragment from the 12S rRNA gene widely used in fish eDNA metabarcoding studies, and then compared eDNA metabarcoding data from water samples with traditional sampling using nets. For the taxonomic discriminatory power analysis, we used a specifically curated reference dataset consisting of 373 sequences from 264 neotropical fish species (including 47 newly generated sequences) to perform a genetic distance-based analysis of the amplicons targeted by the MiFish primer set. We obtained an optimum delimitation threshold value of 0.5% due to lowest cumulative errors. The barcoding gap analysis revealed only a 50.38% success rate in species recovery (133/264), highlighting a poor taxonomic resolution from the targeted amplicon. To evaluate the empirical performance of this amplicon for biomonitoring, we assessed fish biodiversity using eDNA metabarcoding from water samples collected from the Amazon (Adolpho Ducke Forest Reserve and two additional locations outside the Reserve). From a total of 84 identified Molecular Operational Taxonomic Units (MOTUs), only four could be assigned to species level using a fixed threshold. Measures of α-diversity analyses within the Reserve showed similar patterns in each site between the number of MOTUs (eDNA dataset) and species (netting data) found. However, β-diversity revealed contrasting patterns between the methods. We therefore suggest that a new approach is needed, underpinned by sound taxonomic knowledge, and a more thorough evaluation of better molecular identification procedures such as multi-marker metabarcoding approaches and tailor-made (i.e., order-specific) taxonomic delimitation thresholds.

The eDNA component of the study was conducted in six different sites located in the Brazilian Amazon, with four sites inserted inside the Adolpho Ducke Forest Reserve and two outside. 

For the eDNA metabarcoding, from each of the four streams within Ducke Reserve, three water replicates were taken from three positions (from the left bank, middle and right bank) along a transect at 0m, 25m, and 50m using 500 ml water bottles resulting in a total of nine replicates per stream (9 × 500 ml). Two water‐sample replicates were collected at site A and three at site C (no netting was performed at these sites). At the start of each sampling period, a field blank was collected, totalling four field blanks overall. Samples were collected prior to the start of the netting sampling and disposable collection bottles and syringes were used to avoid the risk of contamination. Water samples were manually filtered using a syringe and Sterivex enclosed filters (0.45μm, Merck Millipore) and kept cool until transport to the UK.

In total, 51 samples were analyzed (including 41 water samples, four field blanks, two extraction blanks, four PCR negative controls). The eDNA extraction, amplification of the 12S rRNA fragment using the MiFish primer set, and library preparation (Illumina MiSeq ‐ V2 2 × 150 bp kit) were conducted following the procedure described in Sales et al. (2021) and details are provided in the referred manuscript.

This dataset contains the raw data generated in one Illumina MiSeq run (V2,  2 × 150 bp kit), following the procedures described in the methods section.