Discussion

The fish communities of our sampling sites along the Danube and near the mouths of its main tributaries are well known both in terms of the fish species list and the assemblage structure (Eros et al., 2017; Kottelat & Freyhof, 2007; Sommerwerk et al., 2009); thus, these communities are useful for testing the effectiveness of an eDNA metabarcoding strategy. From a total of 86 taxa detected during our study, only five were UNKTaxa in the Danube catchment (Kottelat & Freyhof, 2007; Sommerwerk et al., 2009). For most of these taxa, the main explanation is probably a misassignment of the detected sequences in relation to insufficient knowledge of their regional haplotype variability. Richardsonius balteatus , a North American species, Barbus meridionalis , present in rivers draining to the northwestern Mediterranean basin, andEsox cisalpinus occurring in central and northern Italy (Kottelat & Freyhof, 2007) are species whose teleosequences are close to those ofSqualius cephalus , Barbus barbus and Esox lucius,respectively. Oncorhynchus clarkii and Oncorhynchus masou , two salmonid species inhabiting the northern Pacific Ocean, may also have been confused with 0nc_Myk, but they have also been introduced into European fish farms (Crawford & Muir, 2008), and hybridization with other Salmonid species is conceivable (Chevassus, 1979). The development of a more comprehensive local reference database would reduce this risk of misassignment.
The WASTaxa category of taxa was composed mainly of food fish according to the eDNA present in urban wastewater. Most of these taxa were detected at only three sites: immediately downstream of the wastewater discharge point of the city of Vienna, on the Argès River and on the Russemski Lom River. The latter two rivers are known to receive insufficiently treated municipal wastewater (Frincu, 2021; Kirschner et al., 2021). eDNA released into the river from wastewater treatment plants can lead to false-positive detection results, and a good knowledge of the regional fauna is needed to identify them. Notably, the detection of marine food fish is a clear sign of local pollution and can be incorporated as a criterion for future bioassessment methods based on eDNA samples (Pont et al., 2021). Two other taxa (0nc_Myk, Sal_spp) are also known as food fish and farmed fish (https://www.helgilibrary.com/indicators/fish-consumption-per-capita/austria/), but they are also regularly present in the Upper Danube and its tributaries, mainly due to stocking (Stankovic, Crivelli, & Snoj, 2015). Therefore, the presence of their eDNA must be interpreted with caution when detected in a water body that does not correspond to one of their known habitats.
A total of 60 taxa known to occur in the Danube River catchment (KNWTaxa) were detected. In addition to the 48 taxa assigned at the species level, the 12 taxa assigned at a higher taxonomic level corresponded to a potential of 26 well-known Danubian species, giving a maximum number of 74 species detected. This value was comparable to the total of 71 species caught in the TEF survey conducted in the same period (Bammer et al., 2021). When considering only the 18 sites sampled with both TEF and eDNA, all the species caught by using TEF were detected by using eDNA except four (Clupeonella cultriventris, Eudontomyzon danfordi, Eudontomyzon mariae, Neogobius eurycephalus ), but they were not recorded in our DNA reference database. Six of the eight taxa (Aci_gue, Aci_rut, Aci_ste, Bar_car, Ben_sp, Rom_ura) detected only by using eDNA were benthic species (Kottelat & Freyhof, 2007) inhabiting mainly the Danube itself or its coarse-bottomed tributaries. Similarly, the higher taxonomic richness obtained by using eDNA confirmed the ability of this method to be representative of all fish fauna, especially in deep rivers where a single traditional sampling technique does not allow sampling of the whole river section (Eros et al., 2017). Our results highlight the effectiveness of our integrative sampling strategy in space (the whole section of the river) and time (approximately half an hour) as well as the performance of the teleo primer, even if its discriminating power for some species is limited. For the latter, the analysis of another marker in parallel, such as MiFish, can allow more species to be discriminated (Polanco et al., 2021).
One of the most original aspects of this study is the strong correlation between teleo-eDNA concentrations and fish abundance estimated by using TEF at 18 common sites. The efficiency of eDNA qPCR data to correctly estimate taxa-specific abundance is well documented (Rourke et al., 2021), but the estimation of the total fish abundance from the total fish eDNA concentration (primer qPCR analysis) has been tested only in an estuarine environment at three sites only a few kilometres apart (van Bleijswijk et al., 2020). Here, we demonstrate the capability of eDNA metabarcoding to estimate the total absolute abundance of fish at distant sites, i.e., independent of their eDNA contents, and in a large range of river sizes. The intensity of the correlation between the teleo-eDNA concentration and fish abundance is comparable to results obtained in species-specific qPCR studies in natural environments (Yates et al., 2019). The difference in correlation intensity with fish abundance observed when the eDNA concentration is expressed as density or biomass should be viewed with caution, as no significant effect of the fish abundance metric was found (Yates et al., 2019). The ratios of fish species-specific read counts over the total read count of a sample multiplied by the teleo-eDNA concentration measured with qPCR (van Bleijswijk et al., 2020) were significantly correlated with the fish species abundance obtained by using TEF. This correlation was higher when all sites were pooled, which highlights the agreement between the two methods for all the species and the importance of uncertainties associated with the site scale with both eDNA and TEF.
The very high values of the co-inertia criteria also demonstrate that the descriptions of fish community structures obtained with the TEF (abundance per ha) and eDNA methods (taxa-specific DNA copy numbers per litre) were quite similar. The distribution of species along the entire Danube River obtained by using eDNA was consistent with previous knowledge (Eros et al., 2017) but with a lower between-site variability. For example, Aci_rut, a resident sturgeon species, was regularly detected downstream of the first 1000 km of the river by using eDNA, whereas no or few individuals were captured by using traditional methods (Bammer et al., 2021; Eros et al., 2017). The anadromous taxa Alo_spp (Alosa immaculata / A. tanaica ) was detected by using eDNA in almost all the sites located downstream of the Iron Gate dams that are known to limit their upstream migration (Sommerwerk et al., 2009). In addition, the detection of Alo_spp 12 km upstream of Iron Gate I dam (KM 1908) is consistent with previous captures of Alosa tanaicaindividuals upstream of Iron Gate II (M. Lenhardt, pers. comm.).
Nevertheless, eDNA is only an indirect estimator of organism abundance and is influenced by many physiological processes and environmental conditions, and the uncertainties associated with all factors affecting eDNA concentration in the environment are high (Rourke et al., 2021). eDNA cannot be expected to provide a highly accurate quantification of the fish populations as needed for precise fish stock estimations in fisheries (Boivin-Delisle et al., 2021; Rourke et al., 2021; Yates, Cristescu, et al., 2021). For such a purpose, recent technical options could provide a good alternative (Hoshino et al., 2021; Sato et al., 2021; Taylor M. Wilcox et al., 2020; Ushio et al., 2018). However, it must also be considered that most conventional fish sampling methods are associated with many biases and high uncertainties, especially in large water bodies where the spatial representativeness of samples is limited and multiple methods must be used (Eros et al., 2017; Zajiceke & Wolter, 2018). For most biomonitoring purposes, a rough estimation of absolute fish abundance is sufficient, as the main objective is to compare fish assemblages on a large scale or to detect long-term variability in relation to changes in anthropogenic disturbances.
An additional benefit of quantifying total fish eDNA by qPCR is to optimize sampling effort. Our NLME models showed that the species richness was underestimated when the amount of teleo-eDNA extracted from a sample was below a threshold of 0.65.106 eDNA copies. Although several authors have recognized the importance of this parameter (Shu, Ludwig, & Peng, 2020; Wang et al., 2021), to our knowledge, no studies have quantified its influence. In addition, our results demonstrated the significant influence of river size on the concentration of teleo-eDNA per litre, with values 10 to 100 times lower in larger rivers. This can be due to different processes, e.g., dilution of eDNA with increasing river depth, as most fish species are confined to the river bottom or shoreline, or the decreased abundance of fish in large rivers compared to small rivers. Further research is needed to better understand the processes that explain such a pattern. As the quantity of teleo-eDNA extracted depends on both its concentration per litre and the water volume sampled, the water volume needed to extract an amount of eDNA over the threshold of 0.65x106 eDNA copies is approximately 40 litres for large rivers but only a few litres for smaller rivers. The volume of water to be sampled is the main issue in many studies, with values ranging from less than a litre to 68 L (Cantera et al., 2019; Civade et al., 2016; Doi et al., 2017), but no general guidelines have been established (Shu et al., 2020; Wang et al., 2021). This study highlights that river size is one of the main factors that influences the minimum water volume to be sampled. Nevertheless, this result is only valid in the context of our spatial and temporal integrative sampling strategy: the total volume collected must be sufficient to allow the collection of eDNA from the entire river section.
In conclusion, our results show that the combination of qPCR analysis to estimate the total concentration eDNA amplified by the “teleo” primer, an eDNA metabarcoding workflow with a high number of technical replicates, and an integrative sampling strategy allows a correct estimation of species diversity and delivers a good proxy of absolute species abundance (based on taxa-specific DNA copy numbers per litre). Our approach is not appropriate if accurate abundance estimation is required, such as in intensively managed fisheries. However, we consider it sufficient for most biomonitoring and bioassessment purposes, especially given the limited effectiveness of conventional fish sampling methods in most aquatic ecosystems. The efficiency of our procedure needs to be tested in ponds and lakes, estuaries, and marine environments. Our results should inspire a more quantitative approach to aquatic community analysis using eDNA methods.