The main objective of this work was to develop and validate a robust and reliable "from-benchtop-to-desktop" metabarcoding workflow to investigate the diet of invertebrate-eaters. We applied our workflow to faecal DNA samples of an invertebrate-eating fish species. A fragment of the cytochrome c oxidase I (COI) gene was amplified by combining two minibarcoding primer sets to maximize the taxonomic coverage. Amplicons were sequenced by an Illumina MiSeq platform. We developed a filtering approach based on a series of nonarbitrary thresholds established from control samples and from molecular replicates to address the elimination of cross-contamination, PCR/sequencing errors and mistagging artefacts. This resulted in a conservative and informative metabarcoding data set. We developed a taxonomic assignment procedure that combines different approaches and that allowed the identification of ~75% of invertebrate COI variants to the species level. Moreover, based on the diversity of the variants, we introduced a semiquantitative statistic in our diet study, the minimum number of individuals, which is based on the number of distinct variants in each sample. The metabarcoding approach described in this article may guide future diet studies that aim to produce robust data sets associated with a fine and accurate identification of prey items.

The development of DNA barcoding from faeces represents a promising method for animal diet analysis. However, current studies mainly rely on prior knowledge of prey diversity for a specific predator rather than on a range of its potential prey species. Considering that the feeding behaviour of teleosts may evolve with their environment, it could prove difficult to establish an exhaustive listing of their prey. In this article, we extend the DNA barcoding approach to diet analysis to allow the inclusion of a wide taxonomic range of potential prey items. Thirty-four ecological clade-specific primer sets were designed to cover a large proportion of prey species found in European river ecosystems. Selected primers sets were tested on isolated animal, algal or plant tissues and thereafter on fish faeces using nested PCR to increase DNA detection sensitivity. The PCR products were sequenced and analysed to confirm the identity of the taxa and to validate the method. The methodology developed here was applied to a diet analysis of three freshwater cyprinid species that are assumed to have similar feeding behaviour [Chondrostoma toxostoma toxostoma (Vallot 1837), Chondrostoma nasus nasus (Linnaeus, 1758) and Barbus barbus, (Linneaus 1758)]. These three species were sampled in four different hydrographic basins. Principal Component Analysis based on prey proportions identified distinct perilithon grazer and benthophagous behaviours. Furthermore, our results were consistent with the available literature on feeding behaviour in these fish. The simplicity of the PCR-based method and its potential generalization to other freshwater organisms may open new perspectives in food web ecology.

In diet metabarcoding analyses, insufficient taxonomic coverage of PCR primer sets generates false negatives that may dramatically distort biodiversity estimates. In this paper, we investigated the taxonomic coverage and complementarity of three cytochrome c oxidase subunit I gene (COI) primer sets based on in silico analyses and we conducted an in vivo evaluation using fecal and spider web samples from different invertivores, environments, and geographic locations. Our results underline the lack of predictability of both the coverage and complementarity of individual primer sets: (a) sharp discrepancies exist observed between in silico and in vivo analyses (to the detriment of in silico analyses); (b) both coverage and complementarity depend greatly on the predator and on the taxonomic level at which preys are considered; (c) primer sets’ complementarity is the greatest at fine taxonomic levels (molecular operational taxonomic units [MOTUs] and variants). We then formalized the “one‐locus‐several‐primer‐sets” (OLSP) strategy, that is, the use of several primer sets that target the same locus (here the first part of the COI gene) and the same group of taxa (here invertebrates). The proximal aim of the OLSP strategy is to minimize false negatives by increasing total coverage through multiple primer sets. We illustrate that the OLSP strategy is especially relevant from this perspective since distinct variants within the same MOTUs were not equally detected across all primer sets. Furthermore, the OLSP strategy produces largely overlapping and comparable sequences, which cannot be achieved when targeting different loci. This facilitates the use of haplotypic diversity information contained within metabarcoding datasets, for example, for phylogeography and finer analyses of prey–predator interactions.