The sampling site was located at the Mulde weir in Dessau (Germany, 51°49’56.2”N, 12°15’05.1”E). The river Mulde is a tributary of the Elbe system with an average effluent at the sampling site of 62.7 m³/s in April (2012–2018; FIS FGG Elbe). From the complete stream system up to 34 fish species are reported (Geisler 2001, MULE fish report 2014), which is close to the total number of 50 fish species reported for the German federal state of Saxony-Anhalt (Kammerad et al. 2020). Amongst these are endangered and strictly protected fish as well as diadromous and invasive species. In accordance with the WFD, a fish ladder was built in 2017 in the stream to surpass the 2.4 m weir and to allow for unimpeded migration of organisms, in particular fish.

We collected 18 water samples on a single day in April 2019 over a stretch of 2 km: 4 samples each were collected 1 km upstream of the weir (location S1), directly upstream (S2) and directly downstream of the fish ladder (S3), and 1 km downstream of the weir (S4). Additionally, two samples were taken directly in the fish ladder itself (S4). For each sample, 1 L of water was collected from the surface in a sterile plastic bottle. To prevent cross-contamination, sterile laboratory gloves were changed between samples. All water samples were filtered on site to avoid contamination and ease the transportation. Open MCE (mixed cellulose ester membrane) filters with a 0.45 µm pore size (diameter 47 mm, Nalgene) were used for the filtration. The filters were handled with sterile forceps (previously bleached with 4%–5% sodium hypochlorite, rinsed with distilled water afterward) and gloves. Both forceps and gloves were changed between each sample. An electric vacuum pump, a disposable funnel filter, and a filter flask were installed for filtering the water. Both the funnel filter and filter flask were used for all samples, since they do not come in contact with the filter membrane. As field blanks, a total of two blank filters were placed on the filter flask, the pump was switched on and the filters were exposed to air for 20 seconds. The filters were transferred to 1.5 mL Eppendorf tubes filled with 96% ethanol, kept at 4 °C during transport and overnight, and then stored at -20 °C until DNA extraction.

All laboratory steps were conducted under sterile conditions in a dedicated sterile laboratory (UV light, sterile benches, overalls, gloves, and face masks). The filters were dried separately in sterile petri dishes and covered with aluminum foil overnight. Afterwards the filters were torn into pieces using sterile forceps and transferred into new 1.5 mL Eppendorf tubes. Subsequently, filters were eluted in 600 µL TNES-Buffer with 10 µL Proteinase K (300 U/mL, 7BioScience, Neuenburg am Rhein, Germany) and incubated at 55 °C and 1000 rpm for three hours on an Eppendorf ThermoMixer C (Eppendorf AG, Hamburg, Germany). DNA was extracted from the filters following an adapted salt precipitation protocol (Weiss and Leese 2016), eluted in 50 µL PCR-grade water, and stored overnight at 4 °C. Next, 0.5 µL RNase A (10 mg/mL) was added to each sample and incubated for 30 minutes at 37 °C on an Eppendorf ThermoMixer C. Subsequently, samples were purified using the Qiagen MinElute DNeasy Blood & Tissue Kit (Hilden, Germany), following the manufacturer’s protocol. Samples were eluted in 30 μL PCR-grade water.

A two-step PCR approach was applied for amplifying the extracted DNA. In the first PCR, the vertebrate tele02 primers (Taberlet et al. 2018) were used that are optimized for European freshwater fish and amplify a 129–209 bp long 12S gene fragment. In total, 100 first step PCR reactions were run, including 5 replicates per sample as well as 8 negative PCR controls and 2 field blanks. The PCR reaction volume was 50 µL and consisted of 21 µL PCR-grade water, 25 µL Multiplex Mastermix (Multiplex PCR Plus Kit, Qiagen, Hilden, Germany), 1 µL of each the tele02 forward and reverse primer (10 µM), and 2 µL DNA template. The first PCR step was carried out at 95 °C for 5 minutes followed by 35 cycles with 94 °C for 30 seconds, 52 °C for 90 seconds, and 72 °C for 90 seconds. The final elongation was carried out at 68 °C for 10 minutes. After the first PCR, all five replicates per sample were pooled together. For the second PCR step, a universal tagging primer set was used (Buchner et al. under review). A total of 52 second-step PCR reactions were run using two PCR replicates per sample, 4 first-step negative controls, 4 second-step negative controls, and 2 field blanks. The PCR mix per sample contained 19 µL of PCR-grade water, 25 µL of Multiplex Mix, 2 µL combined primer (10 µM), and 4 µL PCR product from the first PCR. PCR conditions were 95 °C for 5 minutes followed by 10 cycles at 94 °C for 30 seconds, 62 °C for 90 seconds, and 72 °C for 90 seconds. The final elongation was carried out at 68 °C for 10 minutes. Following the second-step PCR, the PCR products were visualized on a 1% agarose gel to evaluate the amplification success. The samples were subsequently normalized to 25 ng per sample, using a SequalPrep Normalization Plate (Applied Biosystems, Foster City, CA, USA) following the manufacturer’s protocol. Subsequently, the normalized samples were pooled into one library. After library-pooling, the samples were concentrated using a NucleoSpin Gel and PCR Clean-up kit (Machery Nagel, Düren, Germany) following the manufacturer’s protocol. The final elution volume of the library was 22 µL. The samples were then analyzed using a Fragment Analyzer (High Sensitivity NGS Fragment Analysis Kit; Advanced Analytical, Ankeny, USA) to check for potential primer dimers and co-amplification and quantify the DNA concentration of the library. Primer dimers were removed by extracting PCR products using two lanes (10 µL each) of an E-Gel Power Snap Electrophoresis Device (ThermoFisher Scientific, Germany). This resulting library was sequenced on a MiSeq using the v2 250 bp PE Illumina kit at CeGaT (Tübingen, Germany).

Raw reads for both libraries were received as demultiplexed fastq files. The quality of the raw reads was checked using FastQC (Andrews 2010). The sequencing data was processed using a pre-release version of the graphical-user interface pipeline MetaProcessor (available at https://github.com/TillMacher/MetaProcessor). Paired-end reads were merged using VSEARCH version 2.11.1 (Rognes et al. 2016), allowing for up to 25% differences between merged pairs and a minimum overlap of 5 bp. Afterwards, primers were trimmed with cutadapt version 2.8 (Martin 2011), using the linked adapter option without anchoring. Reads were then filtered by length (±10 bp threshold for the amplified 129–209 bp tele02 target fragment) and by maximum expected error (maxee = 1), using VSEARCH. The filtered reads were dereplicated and singletons and chimeras were removed with VSEARCH. All reads were then pooled using a custom python script and again dereplicated (MetaProcessor: v_derep_singletons_uchime.py). Operational taxonomic units (OTUs) were obtained by a 97% similarity clustering and the seeding sequences were extracted as representative OTU sequences. The OTUs were remapped (usearch_global function, 97% similarity) to the individual sample files to create the read table. The read table was filtered by column (read abundance threshold: >0.01% of reads to keep the OTU) and then by row (OTU must be present in at least one of the samples). BLAST searches (web BLAST, blastn suite, nt database, blastn program) were performed against the National Center for Biotechnology Information (NCBI) database (access date: 17.10.2020). The results were downloaded in xml2 format and processed using a custom python script (MetaProcessor: xml2_converter.py). Here, the taxon ID and BLAST similarity columns were fetched from the xml2 file (Suppl. material 1: Table S1 sheet “Raw hits”) and the respective taxonomy was downloaded from the NCBI server (Suppl. material 1: Table S1 “Taxonomy added’’). The BLAST results were subsequently filtered in three steps. First, only the hit with the highest similarity was kept and duplicate hits were removed. When two or more different taxon names were found, all of them were kept. Subsequently, the hit table was filtered according to the thresholds described in JAMP (https://github.com/VascoElbrecht/JAMP), with a >= 98% cutoff threshold to retain species level, >= 95% for genus, >= 90% for family, >= 85% for order and below 85% for class level (Suppl. material 1: Table S1 “JAMP filtering”). Subsequently, all remaining hits of one OTU were trimmed to their first shared taxonomic rank. Remaining duplicates (i.e., hits of one OTU that share the same taxonomy after the filtering) were dereplicated. Thus, each OTU was assigned to one taxonomic hit in the final taxonomy table (Suppl. material 1: Table S1 “JAMP hit”). Finally, OTUs were matched with the read table and OTUs without assigned taxonomy after BLAST searches were discarded.

Both, the taxonomy and read table file were converted to the TaXon table format (Suppl. material 2: Table S2) for downstream analyses in TaxonTableTools v1.2.4 (Macher et al. 2021). The separately sequenced PCR and field replicates were analyzed using the replicates analysis tools (correlation analyses and shared OTUs) and were subsequently merged. Furthermore, per OTU the sum of reads present in negative controls were subtracted from each sample and only OTUs assigned to the phylum Chordata were kept for the downstream analyses (Suppl. material 3: Table S3). Spearman rank correlation analysis was performed for i) read numbers and ii) OTU ranks between replicates to test for stochastic deviations. The read proportions, number of OTUs and number of unique species for each class were calculated. A Jaccard distance-based non-metric multidimensional scaling analysis (NMDS) was conducted to test if site effects between the five sampling locations were present or if all samples can be treated as individual field replicates (NMDS settings: 8000 iterations and 400 different initializations). Three incidence data-based rarefaction and extrapolation analyses were performed to calculate the effect of field replicates on the number of obtained species using the iNEXT online calculation tool (https://chao.shinyapps.io/iNEXTOnline/; Hsieh et al. 2016) with 1000 bootstrap repetitions. Here, random sub-samples were drawn from all 18 field replicates and the number of observed fish/lamprey, bird, and mammal species counts were assessed separately. Additionally, an occupancy plot was calculated in TaxonTableTools to investigate the relative occurrence of each species across all replicates. The plot was subsequently adjusted in Inkscape to add an order-specific color code.

Using eDNA metabarcoding, we successfully detected 25 fish species known to occur in the river Mulde and, further, even 50% of all fish species native to Saxony-Anhalt. Most fish species belonged to the order of Cypriniformes (66% of all species), which was expected since they are the dominant group in Central European rivers (Freyhof and Brooks 2011). The species that stood out in terms of read abundance (57.7% of all reads) was the common dace (Leuciscus leuciscus), followed by the Eurasian ruffe (Gymnocephalus cernua, 4%), and the common bream (Abramis brama, 4%). Quantitative interpretations of read counts and biomass or specimens abundance have been reported for fish (Hänfling et al. 2016; Ushio et al. 2018b; Salter et al. 2019; Muri et al. 2020) but can be prone to several sources of bias. In our study, the sampling event took place during the spawning time of various fish species in spring. This can lead to a potential inflation in the number of eDNA molecules of certain species that for example release their eggs and sperm into the open water, such as the common dace (Mills 1981). Furthermore, we cannot rule out a primer-specific bias that in- or deflates read counts of certain species. Thus, we here omitted correlations of read counts to specimen abundance or biomass and merely focused on species occurrence.

However, not all OTUs were successfully assigned to species level. We found multiple taxa where the 12S marker resolution was too low to distinguish between species and instead two species with identical similarity score were assigned. We manually checked these cases and found several OTUs for which both potential species were reported from the Mulde. For these ambiguous taxonomies we chose a strict approach and counted those cases as one entry. For example, we found the crucian carp and goldfish (Carassius carassius and C. auratus), where the crucian carp is the ancestry species of the domestic goldfish (Chen et al. 2020). Other closely related species we found are the white bream (Blicca bjoerkna) and vimba bream (Vimba vimba), the asp (Leuciscus aspius) and common bleak (Alburnus alburnus), and the invasive bighead and silver carp (Hypophthalmichthys nobilis and H. molitrix). Furthermore, the record of the Macedonian vimba (Vimba melanops) was puzzling, since it does not occur in Germany. We suggest that this hit resembles most likely a vimba bream (Vimba vimba), as we found this native species in our dataset and both are closely related (Hänfling et al. 2009), which may impact the taxonomic assignment. These findings confirmed that the fish-specific tele02 marker is not suitable to distinguish all Central European fish at species level.

While most studies discard all non-target sequences (e.g., Evans et al. 2017; Li et al. 2018; Harper et al. 2019; Sales et al. 2020), we explicitly explored the legitimacy of the detected species and discuss whether they can inform other biomonitoring or species conservation activities. We here used the tele02 primer pair that is known to amplify DNA of vertebrate species other than fish (Mariani et al. 2021). First, we could show that many vertebrate species besides fish were found as bycatch in our samples. We were able to detect a notable 22.2% and 7.4% of the whole native mammal and breeding bird fauna reported from Saxony-Anhalt, respectively. While only a minority of the detected species are water-bound or semi-aquatic, the majority inhabit agricultural, forest, and urban habitats, which accompany large parts of the upstream areas of the river Mulde. All organisms depend on water as a drinking source, which makes streams a sink for eDNA signals, transporting them downstream. The most represented group of the vertebrate species bycatch in terms of read proportions and species richness were mammals. The most represented order within the mammals was rodents (Rodentia). Here, high read counts were assigned to the semi-aquatic Eurasian beaver (Castor fiber), which is reported to inhabit the river Mulde (German national FFH report, 2019). Furthermore, several terrestrial rodents were found, which often inhabit agricultural and urban environments, such as the striped field mouse (Apodermus agrarius) or the Eurasian harvest mouse (Micromys minutus). Four species of even-toed ungulates (Artiodactyla) were detected: Cattle (Bos taurus) are livestock and graze on fields near the river. Roe deer (Capreolus capreolus) and red deer (Cervus elaphus) are known to be good swimmers and can easily cross rivers to reach new feeding grounds and thus release traces into the water. Three carnivora species were detected with eDNA. The putative detection of Canis lupus is most likely explained by the detection of domestic dogs (Canis lupus familiaris), which cannot be distinguished from one another based in the analyzed 12S region. However, since wolf populations have significantly increased over the last decades in central Europe (Chapron et al. 2014) and wolves have been reported from the area of the sampling site (LAU Saxony-Anhalt wolf observation report 2020; J. Arle pers. obs.), a detection of a wild wolf cannot be ruled out. The two other detected carnivore species are the beech marten (Martes foina), a generalist and adaptable species inhabiting open areas and forests, and the invasive raccoon (Procyon lotor), which inhabits forests or urban areas and is a good swimmer that prefers freshwater associated habitats. Furthermore, three bat species were found, i.e., the Daubenton’s bat (Myotis daubentonii) and a pipistrelle species (either Pipistrellus pipistrellus or P. pygmaeus). The Daubenton’s bat relies on clean streams or lakes and hunts insects directly over the water surface (Vesterinen et al. 2013, 2016), which makes it very likely to introduce DNA traces into the water by dropping hair, skin, saliva, urine, and feces. The two detected pipistrelle species are closely related and were not distinguishable with the 12S marker. Birds, however, were generally less represented in both read proportions and species richness compared to mammals. An initial observation was that the detected birds are rather common species that occur in high numbers in the area, compared to the detected mammals. Several aquatic and marsh birds were detected, such as the mallard (Anas platyrhynchos), the graylag goose (Anser anser), the mute swan (Cygnus olor), and the common crane (Grus grus). While the first three species are common waterfowl in Germany all year round, the common crane is a migratory bird. Its detection falls directly in the spring migration, when large flocks of common cranes travel northwards, which makes a detection with eDNA very likely. Besides the waterfowl, most detected species belonged to passerine birds. Two puzzling species were detected that are not present in Germany: the pine bunting (Emberiza leucocephalos) and the slaty bunting (Schoeniclus siemsseni). Here, the most likely explanation is that the 12S marker is not suitable to identify them at species level and distinguish them from other bunting species that are inhabiting Germany and are abundant in the area, such as the common reed bunting (Emberiza schoeniclus) or the yellowhammer (Emberiza citrinella). Furthermore, no amphibians or reptiles were found in the dataset. Particularly the absence of amphibians was notable since at least frog species of the genus Rana and toad species of the genus Bufo are commonly occurring in streams and ponds in Central Europe. Since the detection of amphibians is possible with the here used tele02 primer (Mariani et al. 2021) the river Mulde is most likely not a suitable habitat for amphibians, especially during the reproductive season, which falls into the time of the sampling event.

We found no effect of sampling distance on fish species detection. Thus, although samples were collected at five distinct locations of the river Mulde, the 18 collected samples can be considered as individual field replicates rather than 2–4 specific replicates of 5 sites. The lack of a spatial signal is, on the one hand, not unexpected considering that sampling sites were max. 2 km apart, which is well in the range of reported transport distances of eDNA (Deiner and Altermatt 2014; Shogren et al. 2017; Nukazawa et al. 2018). On the other hand, source, state, transport, and fate of eDNA is anything but simple (Barnes and Turner 2016). While eDNA molecules are transported downstream in general, they are influenced by shedding, retention, and resuspension processes along the way (Shogren et al. 2017). Also, location and density of populations thus can greatly influence the detectability (Carraro, Stauffer, and Altermatt 2021). Community signal inferred via eDNA can thus be very site-specific (Cantera et al. 2019). Besides the spatial aspects, sampling time may be even more important in streams, as eDNA concentration can be increased for several taxa due to e.g., seasonal events such as spawning (Wacker et al. 2019) and migration (Thalinger et al. 2019). Also, water discharge drastically differs among seasons thus leading to different baseline concentrations, suggesting the use of hydrological models in eDNA assessments to increase reliability (Carraro et al. 2018).

Generally, the probability of detecting target DNA when present, i.e., the sensitivity of a method, depends on the concentration and dispersion of target DNA molecules at a site, the sampling design, and the laboratory workflow (Furlan et al. 2016). In previous studies both increased water volume filtered and implementation of field and PCR replicates were found to enhance the sensitivity of eDNA monitoring approaches (Civade et al. 2016; Hänfling et al. 2016; Evans et al. 2017; Beentjes et al. 2019). For example, a previous study on tropical stream fish species suggested that a saturation of tropical stream fish species detection can be reached when sampling 34 to 68 liters per site (Cantera et al. 2019). Our results based on 18 1-L water samples showed that the detection probability of eDNA for non-fish vertebrate species differed substantially among samples. Comparing field samples, we found that fish species richness increased only by 24.8% when considering one versus all 18 samples. A sample volume of 8–10 L can be considered a good tradeoff (sampling effort versus species detection) to monitor the fish fauna of a medium-size river (like the Mulde), since no substantial richness increase was observed by collecting additional 10 L of water. This was different for the detection probability of the non-fish bycatch vertebrate species. For mammals and birds, it increased by 68.9% and 77.3%, respectively, when including 18 field samples. Here, the extrapolation analysis suggested an even further increase of species richness through the collection of additional water samples. While in aquatic organisms such as fish release all their DNA into the surrounding water, only traces of the predominantly terrestrial or aerial bycatch species enter the water leading to a much lower concentration. This expectation is also met when comparing eDNA detection of semi-aquatic bird and mammal species, which were detected in more than 50% of the samples (e.g., mute swan, graylag goose, and Eurasian beaver). Similarly high detection rates, however, were also found for domestic animals living in high abundances in the riparian area of the river (e.g., cattle). Harper et al. (2019) identified similar patterns that terrestrial mammal eDNA signals are weaker and can be detected less frequently than signals from semi-aquatic mammals, using a vertebrate specific primer (Riaz et al. 2011). We also found a high number of human reads in the samples, which are expected in an eDNA assessment in urban environments from various potential sources. Importantly, however, our negative controls did not show many human reads (in our case, a maximum of 4 reads in processed libraries) rendering lab contamination as unlikely. The human traces are likely derived from the original water sample. Nevertheless, low read counts are commonly observed in PCR negative controls and might also originate from laboratory contamination.

While often left aside in studies that focus only on fish biomonitoring, the relevance of detected non-fish bycatch species can be high. This holds true in particular for endangered or protected species that are often difficult to monitor and rely on sighting reports or intensive survey campaigns. Additionally, early reports of invasive species occurrence can also trigger timely management options. For the target taxa, i.e., fish, six of the 25 detected fish/lamprey species are listed as near threatened (European eel), vulnerable (crucian carp, European river lamprey, and burbot), endangered (common barbel), and critically endangered (vimba bream) in the German federal state of Saxony-Anhalt. In the bycatch eDNA data, however, our results detected several mammal species that are classified as protected in Saxony-Anhalt, such as the striped field mouse and yellow-necked mouse (both near threatened), the European beaver and the Daubenton’s bat (both vulnerable), the Eurasian harvest mouse (endangered), and possibly the wolf (critically endangered). Although we were able to detect these endangered species, our findings only provide small insights into the whole vertebrate community, since this study was limited in terms of time coverage (one sampling event) and spatial coverage (2 km stretch of one river). The rarefaction analysis results predicted the detection of more mammal and bird species if more samples were collected. However, it is expected that advances in the standardization and operation of fish eDNA metabarcoding will lead to a higher rate of application in research and regulatory monitoring campaigns in the future. This goes in hand with an increasing amount of available bycatch data that can be analyzed and utilized. With hundreds or thousands of eDNA water samples that are potentially collected each year in countries that apply a nationwide routine monitoring, the coverage of water bodies and different habitats will automatically increase. This opens access to obtain highly resolved spatial and temporal data not only on fish distributions, but also detection patterns of bycatch species. The obtained data could be directly collected in online biodiversity databases and used for more comprehensive insights into vertebrate species occurrence and distribution. The additionally acquired data would then also be available for conservation planning and management and could help to increase the extent and accuracy of regional red lists and lead to a better intercalibration with the international red list. This accounts particularly for conservation monitoring under the EU Birds Directive (Directive 2009/147/EC, 2009), the “EU Regulation 1143/2014 on Invasive Alien Species” or the EU Habitats Directive (Council Directive 92/43/EEC, 1992), where data is generally hard to obtain and striking deficits in the monitoring coverage are known. For example, data on distribution and population sizes of the bird fauna is available in great detail, but observations are often conducted on a non-standardized, voluntary basis. For mammals, however, routine monitoring campaigns are even more scarce, since they are costly and time consuming. Here, the fish eDNA metabarcoding data could provide a notable increase of data points that can be sampled and analyzed under standardized conditions and can be evaluated by experts. The potential of obtaining new, additional information on terrestrial species, in particular elusive, rare or protected species without additional costs is immense and may also stimulate major international conservation initiatives currently developed in the context of the post-2020 CBD-framework.

Nevertheless, the reports of non-target species from fish eDNA metabarcoding have to be interpreted with particular caution. Environmental DNA metabarcoding comes with several challenges that can lead to both false negative and false positive identifications (Barnes and Turner 2016). This accounts particularly for terrestrial and aerial species, which are only temporarily interacting with the water and leave only marginal traces.

We also detected species that are generally unlikely to inhabit the catchment and thus likely represent a false-positive result. Here, potential sources are that the marker resolution is too low to distinguish species, the detected eDNA was already degraded, or the reference sequences are incorrectly labeled. But also, introduction of eDNA via effluents from sewage plants or other influxes can falsify the picture of the species distribution (Yamamoto et al. 2016). Particularly false positive signals must be flagged to avoid biased distribution patterns, when they can be identified as such. It also has to be considered that commonly used fish primers, such as the MiFish (Miya et al. 2015) and tele02 primers, are optimized for fish and discriminate the amplification of other taxa, which will most-likely lead to a lower detection rate compared to fish species. To compensate for the primer bias, universal vertebrate primers (e.g., Riaz et al. 2011) could be used, which would allow to monitor fish, mammal, bird, reptile, and amphibian species at once without additional sampling or laboratory efforts. However, if the main target of the routine biomonitoring remains to detect fish, specific primers might perform better. If the goal is to target groups other than fish, the additional usage of specific primer sets for mammals (MiMammal, Ushio et al. 2017) or birds (MiBird, Ushio et al. 2018) on the same DNA extract is possible, but would come with additional laboratory work and costs. These, however, are small and analysis can be automated (Buchner et al. under review), thus the added value for other specific conservation and management programs can be immense.