The research builds on a monitoring programme initiated by the UK Technical Advisory Group under the European Union Water Framework Directive, which sought a reliable, non-invasive alternative to electrofishing and netting for assessing lake fish communities. Earlier pilot work, including a landmark study of Windermere, England’s largest lake, had established that twenty water samples collected at equidistant points around a lake’s perimeter were sufficient to detect the vast majority of fish species and to produce ecologically meaningful estimates of relative abundance. What remained unclear was how much of that effort was truly necessary, and whether cutting the number of samples would distort estimates of species richness or community composition.

To answer those questions, the team assembled an unusually comprehensive dataset: 2,134 water samples collected between January 2015 and March 2019 from 101 lakes spanning the full range of British lake types, from Scoat Tarn at just 4.3 hectares to Loch Lomond at nearly 5,159 hectares, and from shallow, nutrient-rich lowland waters to deep, acidic upland tarns. Most sampling took place in winter, between November and March, when British lakes are typically well mixed and free of thermal stratification. Each two-litre sample was a composite of five subsamples drawn along a 100-metre transect, filtered within 24 hours, and processed using a vertebrate-specific 12S ribosomal RNA marker that amplifies a short fragment of fish DNA.

After bioinformatic cleaning, which removed low-frequency reads below a 0.1 percent noise threshold and excluded taxa that the marker could not reliably resolve to species level, the final dataset contained 1,792 successfully sequenced shoreline samples and recorded 40 fish taxa across the lakes. Individual lakes yielded between 2 and 18 species, with an average of just under 8. The researchers then applied three complementary analytical approaches: rarefaction-based species accumulation curves, statistical estimation of sample coverage, and extensive resampling of the data to simulate reduced sampling effort.

The results were striking. Sample coverage analysis showed that 93 of the 101 lakes achieved at least 95 percent coverage of their fish species with only ten samples, and 96 lakes reached that threshold with eleven. The mean sampling threshold across all lakes was just 5.37 samples. Resampling analysis, which generated hundreds of unique random combinations of samples for each of 82 lakes with fifteen or more samples, confirmed the pattern: at a sample size of ten, fewer than two species went undetected on average in 95 percent of lakes. Community composition estimates were similarly robust, with Bray-Curtis dissimilarity between reduced subsets and the full lake dataset falling below an arbitrary threshold of 0.1 in 95 percent of lakes at fifteen samples.

Perhaps the most surprising finding concerned lake size. Intuitively, one might expect that a vast lake like Loch Lomond would demand far more sampling than a small tarn, because DNA shed by fish in the middle of a large water body may never reach the shore. Yet the analysis found no correlation between lake surface area and the number of samples needed to achieve reliable detection. Instead, the strongest predictor of sampling effort was species richness itself: lakes harbouring more fish species required more samples to capture them all. This means that monitoring programmes can use prior knowledge of a lake’s expected diversity, rather than its size, to plan an efficient survey.

The study also validated a major simplification in field logistics. In the 20 lakes where both shoreline and offshore transects had been sampled, species richness showed no significant difference between the two approaches, and the proportion of total species detected was nearly identical, averaging 87 percent for shoreline samples and 85 percent for offshore ones. Only one species, the invasive topmouth gudgeon Pseudorasbora parva, was unique to offshore samples, while six species were recorded only from shoreline transects. Although ordination analysis revealed small differences in community composition between transect types in some lakes, these were minor compared with differences among lakes. The practical implication is substantial: monitoring teams can skip boats and specialised deep-water sampling equipment entirely during winter, collecting all samples on foot from accessible shoreline points.

The researchers attribute the success of winter shoreline sampling to the physics of temperate lakes in the cold season. With thermal stratification absent, wind and convective mixing circulate water, and the DNA it carries, throughout the lake, transporting genetic material from deeper habitats toward the shore. Cold temperatures also slow the degradation of extracellular DNA, prolonging the window in which it can be detected. This picture is consistent with earlier work showing that in stratified summer conditions, eDNA distribution becomes far more localised, and shoreline sampling can miss deepwater species entirely. The authors are careful to note that their recommendations apply specifically to polymictic conditions in autumn and winter, and may not transfer to summer surveys or to lakes in warmer climates.

There is an important caveat for conservation applications. While halving the sampling effort had little effect on estimates of overall richness and community structure, rare species were disproportionately likely to be missed in reduced datasets. For programmes targeting endangered natives or early-stage invasive species, where a single detection can trigger management action, the authors recommend retaining a minimum of twenty samples per lake. Detection probability in eDNA surveys also depends on factors beyond sample number, including PCR replication, primer choice, extraction protocol and sequencing depth, so laboratories using different workflows may need to recalibrate these thresholds for their own pipelines.

The broader significance of the study lies in its framework rather than any single number. By combining species accumulation curves, coverage estimation and both random and spatially informed resampling, the authors provide a template that other monitoring programmes can apply to their own systems and objectives. Halving the number of samples per lake translates directly into fewer person-hours in the field and lower filtration and molecular processing costs, potentially allowing the same budget to cover twice as many lakes. For a method whose promise is landscape-scale biodiversity monitoring, that kind of cost optimisation may prove as important as the molecular innovations that made eDNA metabarcoding possible in the first place.

Subject of Research: Optimising environmental DNA sampling effort for lake fish biodiversity monitoring

Article Title: Optimising species detection probability and sampling effort in lake fish eDNA surveys

Article References: Sellers, G. S., Jerde, C. L., Harper, L. R., Benucci, M., Di Muri, C., Li, J., Peirson, G., Walsh, K., Hatton-Ellis, T., Duncan, W., Duguid, A., Ottewell, D., Willby, N., Law, A., Bean, C. W., Winfield, I. J., Read, D. S., Handley, L. L., & Hänfling, B. (2024). Optimising species detection probability and sampling effort in lake fish eDNA surveys. Metabarcoding and Metagenomics, 8, Article e104655. https://doi.org/10.3897/mbmg.8.104655

Image Credits: AI Generated

DOI: 10.3897/mbmg.8.104655

Keywords: environmental DNA, eDNA metabarcoding, lake fish, biodiversity monitoring, sampling effort, species richness, Water Framework Directive, shoreline sampling, rarefaction, community composition, Great Britain, conservation