Data analysis
A false positive sequence threshold (0.1%) was applied to the metabarcoding inventory, where taxa recorded within each sample were discarded if their read frequencies fell below this threshold (Hänfling et al., 2016; Lawson Handley at al., 2019). The metabarcoding inventory was then checked for spurious taxa (i.e. taxa that do not occur in the Czech waters or occurred in very few samples), and taxonomic assignments were corrected. Capelin (Mallotus villosus ) was found in one reservoir sample and thus removed from downstream analyses. A number of species level taxonomic assignments lacked credibility and were reassigned to genus level. Sturgeon (Acipenser sturio ) was changed to Acipenser spp. as Siberian sturgeon (A. baerii ) and Danube sturgeon (A. gueldenstaedtii ) were previously captured by conventional methods but these species were missing from our reference database. Vendace (Coregonus albula ) was corrected toCoregonus spp. as maraena whitefish and peled (C. peled ) are more often stocked in Czech waters. Similarly, Arctic char (Salvelinus alpinus ) was corrected to Salvelinus spp. as the closely related brook trout (S. fontinalis ) is more often stocked in Czech waters, and these species cannot be distinguished using the target 12S rRNA fragment. Silver carp (Hypophthalmichthys molitrix ) and bighead carp (H. nobilis ) were merged toHypophthalmichthys spp. as both species are artificially selected, resulting in hybrids, and share the same ecological niches. Finally, river lamprey (Lampetra fluviatilis ) was corrected toLampetra spp. because river lamprey is not present in the studied catchments and sequences likely belong to European brook lamprey (L. planeri ) which does not possess reference sequences. Four species that occur in the studied reservoirs were assigned to family level as they could not be distinguished using the target 12S rRNA fragment. Asp and rudd (Scardinius erythrophthalmus ) reference sequences were identical and assignments therefore corrected to Cyprinidae, and European perch and pikeperch were identical and corrected to Percidae. One species, Burbot (Lota lota ), was completely missed by eDNA metabarcoding having been previously detected in Římov with gillnetting. Excluding Cyprinidae and Percidae, higher taxonomic assignments (i.e. families, including Salmonidae, or higher ranks) were excluded from downstream analyses due to uncertainty and differing species’ ecologies.
Remaining taxa were classified as non-native or according to the International Union for Conservation of Nature (IUCN) Red List categories (Lusk et al., 2017; Pergl et al., 2016). For each reservoir, five abundance categories were created (Table 3) based on previous ichthyological surveys using benthic and pelagic gillnets during the last 15 years (e.g. Blabolil, et al., 2017a). Species presence was also assessed by seining, electrofishing, fyke-nets, trawling (Jůza et al., 2015; Říha et al., 2015), and expert judgment of FishEcU members (www.fishecu.cz).
Data on inflow, meteorological conditions and reservoir operation were used as input variables in the CE-QUAL-W2 model (Wells, 2019) to compute hydrodynamics, age of water along the longitudinal and vertical profile of reservoirs (i.e. a model variable that is used to estimate the residence time, which accumulates at a rate of 1.0 d-1with all new water entering the system having an age of zero, Monsen et al., 2002) and real-time environmental parameters during sampling campaigns. Measured water parameters were plotted using Surfer software (Surfer 9, Golden Software Inc.). The ChlA concentration data were log-transformed to achieve a normal data distribution. The water age data were normalized within the range from 0–1 for each reservoir to evaluate the proportion of fresh inflow water within each reservoir locality.
Site occupancy and regression analyses were conducted in R v.4.0.0 (R Development Core Team, 2020). We calculated sample-based site occupancy estimates for each taxon (i.e. number of positive samples divided by total number of samples, without inflows) as a proxy for relative abundance in the main reservoir body. A Generalized Linear Mixed-effects Model (GLMM) with reservoir identity as a random effect was applied to test the difference in the number of taxa between the main and side tributaries and in the main reservoir. A second GLMM with reservoir identity as a random effect was used to test the difference in the number of taxa between localities, habitats and seasons. Bray–Curtis dissimilarity for samples taken in the same reservoir, season and habitats was computed using the vegdist function in the package vegan (Oksanen et al., 2019). Differences in Bray–Curtis dissimilarity between sites were compared using a GLMM with habitat and sampling season as fixed effects and reservoir identity as a random effect. Spearman’s rank correlation coefficient was used to test the relationship between eDNA-based site occupancy and the taxon score derived from gillnet sampling. Differences in the number of taxa for each reservoir between season and correlation coefficients for different seasons were compared using a paired t-test.
Multivariate analyses were performed in CANOCO 5 software (Šmilauer & Lepš, 2014). We tested whether fish communities in each reservoir (without inflows) were influenced by environmental variables using a Constrained Canonical Correlation Analysis (CCA). Relationships between fish taxa in the main reservoir body and temperature, DO concentration, total ChlA concentration, normalized water age, locality and habitat were tested using a CCA Interactive Forward Selection with the covariable reservoir identity while restricting the permutations. Log-transformed read counts were supplied to a Monte-Carlo permutation test with 9999 permutations using a test of significance on all constrained axes.