Fresh waters cover a very small part of the Earth's surface and hold a disproportionate share of its described species, and they are losing that diversity faster than most terrestrial systems (Dudgeon et al., 2006). Monitoring is the precondition for any response. In Europe the Water Framework Directive obliges member states to classify the ecological status of rivers and lakes from their fish, invertebrates, aquatic plants and algae (European Parliament and Council, 2000), and comparable duties exist elsewhere. The conventional tools, electrofishing, netting, kick sampling and microscopy, are slow, need taxonomic specialists and can damage what they measure.

Environmental DNA, usually shortened to eDNA, offers a different route. Organisms release DNA into the water in mucus, faeces, gametes and shed cells. Ficetola et al. (2008) showed that the American bullfrog could be detected in French ponds from water samples alone. Combining that idea with DNA barcoding (Hebert et al., 2003) and high-throughput sequencing gave metabarcoding, in which a marker shared by a whole group is amplified and all the variants in a sample are sequenced together (Taberlet et al., 2012).

The method has been reviewed several times from the standpoint of molecular ecology (Deiner et al., 2017; Taberlet et al., 2018). The present article takes the standpoint of a monitoring agency or field ecologist deciding how far to rely on it. It considers what metabarcoding does well, where its results can mislead, and which parts of the workflow most need agreed standards.

This is a narrative review of the published literature on eDNA metabarcoding of water samples from rivers, lakes and ponds, with emphasis on fish, amphibians and macroinvertebrates. Single-species assays based on quantitative PCR are mentioned only where they clarify a point about detection. The review does not follow a systematic search protocol and makes no claim to completeness. Sources were chosen because they introduced a method, tested a specific source of bias, compared metabarcoding with established surveys, or addressed its use in regulation.

A metabarcoding survey has five stages, and decisions at each constrain what the final list can mean. Water is collected and its particulate matter captured, normally by filtration through a membrane or an enclosed capsule. DNA is extracted from the filter. A marker region is amplified by PCR with primers designed to match a target group, and sample-specific tags are attached. The pooled products are sequenced. Finally the reads are filtered, grouped and compared with a reference database to assign names (Taberlet et al., 2018).

The choice of marker defines the taxonomic window. For fish the most widely used primers target a short hypervariable stretch of the mitochondrial 12S ribosomal RNA gene, as in the MiFish set (Miya et al., 2015). For invertebrates the cytochrome c oxidase subunit I gene is preferred because it is the standard animal barcode and has by far the largest reference collection, though its variable primer sites make even amplification difficult. No single marker serves all groups, and surveys aiming at whole communities usually run two or more.

Bioinformatic practice has shifted from clustering reads into operational taxonomic units at a fixed similarity threshold towards inferring exact sequence variants with error models such as DADA2 (Callahan et al., 2016). Exact variants are reproducible labels that can be compared directly between studies, which clusters defined within one data set cannot. For monitoring, where results from different years and laboratories must be set side by side, that property matters a good deal.

The first strength is sensitivity. In Danish ponds, lakes and streams, eDNA detected rare and threatened amphibians, fish, mammals, insects and crustaceans at most of the sites where they were known to occur (Thomsen et al., 2012). A large comparison across many sites found that metabarcoding recorded amphibian and fish species at least as often as traditional methods (Valentini et al., 2016). In Windermere, a lake with decades of fish survey data, eDNA metabarcoding recovered nearly all the species known from the lake, and the relative frequency of their sequences reflected their long-term rank abundance (Hänfling et al., 2016).

The second is that sampling is non-destructive. Collecting a few litres of water harms neither the organisms nor the habitat, which matters for protected species and for sites where electrofishing or netting is restricted. The third is breadth: one sample can be analysed for fish, amphibians, molluscs and insects with different primers, and the extract can be stored and analysed again when better markers or reference data appear.

A fourth strength is less often stated. Sequence data are an auditable record. A list produced by a field surveyor rests on that person's identifications, whereas raw reads can be deposited, reprocessed and reassigned by anyone. Field effort is also modest and needs little taxonomic training, which makes dense spatial coverage and volunteer sampling practical. The cost has not disappeared, however. It has moved from the riverbank to the laboratory and the reference library.

Each stage of the workflow introduces its own distortions. Figure 1 pairs the stages with the main bias arising at each, and Table 1 summarises the consequences for monitoring and the remedies in common use.

Standardization does not require that every laboratory use the same kit. It requires that results be comparable and that their reliability can be judged. Goldberg et al. (2016) set out the minimum considerations for eDNA detection studies: explicit sampling design, contamination control in the field and laboratory, negative and positive controls at each stage, and full reporting of methods. Zinger et al. (2019) made a parallel case for metabarcoding, arguing that weak experimental design, with too few replicates and controls, undermines ecological conclusions however deep the sequencing.

For regulatory monitoring four areas need agreement. The first is field practice: volume filtered, number and placement of samples, filter type and preservation. The second is the laboratory: a fixed set of validated markers for each biological quality element, mock communities run alongside samples, and participation in ring tests between laboratories. The third is data: a defined pipeline or defined performance criteria, with raw reads and metadata archived. The fourth is interpretation. Existing biotic indices were calibrated on abundance data from nets and kick samples, and metabarcoding supplies, at best, presence and relative read frequency.

Two routes for the last problem are under discussion. One keeps the existing indices and feeds them presence data from DNA, accepting some loss of information. The other builds new indices directly from sequence data, including taxa never used in conventional assessment (Pawlowski et al., 2018). Hering et al. (2018) reviewed the options for the Water Framework Directive and judged that DNA-based identification could be taken up most readily for fish, with invertebrates and diatoms requiring further work on abundance and on intercalibration with the long-term record.

Several questions remain open. The relation between read frequency and abundance holds in some systems, as in Windermere, and fails in mock community tests; when and why it holds is not well understood. Hydrological models that convert a detection into a probable source area are at an early stage. Most validation has been done in temperate lakes and streams, and large, turbid, tropical rivers with high sediment loads and poorly catalogued faunas present harder conditions for filtration and assignment alike.

There is also a question of continuity. Monitoring series are valuable because they are long, and replacing the method breaks the series. A period in which eDNA and conventional surveys are run side by side at the same sites is costly and is the only sound way to learn how the two relate. Until that overlap exists, metabarcoding is best used as a complement: for screening large numbers of sites, for detecting invasive or protected species early, and for groups that conventional methods sample poorly.

Environmental DNA metabarcoding has earned a place in freshwater monitoring. It is sensitive, harmless to the system sampled, broad in taxonomic reach and open to audit. It is also a chain of steps, each of which can add false negatives or false positives, and its output describes DNA in water, from which the presence of organisms must be inferred. The priorities are well understood: replication and controls reported as a matter of routine, validated markers, complete and curated reference libraries, archived raw data, and indices designed for the kind of information the method yields. With those in place, the technique can extend conventional surveys to more sites and more taxa than field teams could ever reach.

To be declared by the author(s) before publication.

To be declared by the author(s) before publication.