Environmental DNA for species monitoring

Every living organism leaks DNA continuously: sloughed skin and gill epithelium, mucus, gametes, feces, carcasses. Outside the body this material is called environmental DNA, and it survives long enough to be filtered out of a river, a ballast tank or a soil slurry. Polymerase chain reaction, in turn, can find a single target molecule against any background. The whole technology is the distance between those two facts: sampling becomes filtration, species identification becomes amplification, and an animal is registered without being seen, caught or harmed.

Why water carries a readable signal

Most free DNA in water does not float naked. It adsorbs onto suspended particles, sediment and biofilms, which shields it from nucleases, and it is released faster than it is destroyed, so a population keeps refreshing the signal downstream. A water sample is therefore an integrated, time-averaged census of who has been shedding in the catchment — including nocturnal, rare and cryptic species that visual and capture surveys miss systematically. Two readouts exist. Targeted droplet digital PCR partitions an extract into tens of thousands of nanolitre droplets and counts how many amplify, giving near-single-copy sensitivity for one named invader. Metabarcoding instead amplifies a shared marker — 12S for vertebrates, COI for invertebrates, 16S or ITS for microbes and fungi — and sequences the mixture, converting one sample into an inventory of a whole community. Both routes end the same way: sequences are matched against a reference database, so identification is never better than that database.

Degradation sets the clock, transport sets the place

DNA in water is attacked by nucleases, ultraviolet radiation and hydrolysis; in warm, sunlit, alkaline water the detectable signal lasts days rather than months — exactly what an early-warning system wants. In cold, anoxic sediment the same chemistry runs slower and fragments persist for years, so sediment reads history rather than presence and can overturn a “site is clean” claim with arrivals from decades past. Transport blurs space: DNA shed upstream is detected downstream, diluted by tributaries, trapped in eddies. A positive detection is a statement about a drainage area and a time window, not a point coordinate; converting it into “where and how many” requires a hydrological model of dilution and decay, and even then the answer stays coarse.

Detection probability is the real measurement

At single-copy sensitivity, sampling is stochastic: whether a litre of water holds a target fragment is a Poisson event, so replicate filters from one site disagree, and a species can be present yet undetected. False negatives also come from humic inhibitors that survive extraction and from primer mismatches in divergent lineages. False positives are more insidious, because at this sensitivity one stray amplicon — carryover, index misassignment during sequencing, a contaminated reagent — becomes a detection. The honest output of an eDNA survey is therefore not a species list but an occupancy estimate with an explicit detection probability, fitted from repeated surveys and controls, separating “absent” from “not detected”. That statistical layer, not the chemistry, concentrates most of the difficulty and most of the current methodological argument. It is the same estimation problem that passive acoustic monitoring faces, and both methods consume the same upstream resource: authenticated reference sequences.