Comparing eDNA and Transect Methods for Aquatic Biodiversity Assessment in Lakes and Ponds

ABSTRACT

Biodiversity monitoring increasingly relies on molecular methods such as eDNA metabarcoding. However, sound applications have so far been only established for a limited number of taxonomic groups. More information on the strengths and weaknesses of eDNA methods, especially for poorly covered groups, is essential for practical applications to achieve the highest possible reliability. We compared amphibian and Odonata data from eDNA metabarcoding and traditional transect walks on N = 56 plots in 38 water bodies distributed over six extraction sites for building materials in Northwest Germany. The traditional amphibian assessment included visual encounters, dip netting and acoustic detection, while Odonata were assessed through exuviae. In total, both methods detected 8 out of 11 amphibian species, while the remaining three species were detected by eDNA only. We did not find differences in amphibian species numbers per plot, but mean detection probabilities were higher with metabarcoding. In contrast, both methods detected 10 out of 29 Odonata species, while the remaining 19 species were detected by exuviae only. Species numbers per plot were higher for exuviae and only 30% of species were detected with metabarcoding. The species identified by eDNA were those with high abundance, and their detection probabilities were similar to transect walks. The results for amphibians show equal suitability and high complementarity of the compared methods. Metabarcoding detected species more efficiently and therefore offers a suitable protocol for biodiversity monitoring. For Odonata, eDNA metabarcoding showed considerable gaps, implying the need for protocol evaluation and improvement in assessment of ecological communities based on eDNA.

1. Introduction

Biodiversity loss and environmental change increase the need for cost and time‐efficient monitoring methods (Schmeller et al. 2017). New technologies should optimally be easy to apply, faster and deliver equal or even better results than conventional monitoring methods. The use of eDNA metabarcoding as one of these promising new technologies has increased significantly over the last years (Deiner, Yamanaka, and Bernatchez 2021; Veilleux, Misutka, and Glover 2021). Apart from method testing, metabarcoding is already widely used for practical conservation and monitoring purposes (Blackman et al. 2022; Schenekar 2023). Its comprehensive character allows for the generation of large datasets of whole communities across orders, families and taxa with comparatively little time and money (Banerjee et al. 2022; Deiner et al. 2016; Reji Chacko et al. 2023). Targeting threatened species, the non‐invasiveness of eDNA sampling in general can be crucial to enable monitoring at all (Schmidt et al. 2021; Sittenthaler et al. 2023; Thomsen et al. 2012). Nevertheless, there are also limitations that can be disadvantageous, depending on the question being asked. For example, missing information on abundance or demographic structure can be a problem, if the monitoring of population dynamics is in focus instead of pure presence/absence data (Fonseca 2018; Sickel et al. 2023; Skelton, Cauvin, and Hunter 2023). However, since presence/absence data are often sufficient, disadvantages rather tend to arise from methodological uncertainties. Fluctuations in dispersion and persistence of eDNA have been investigated in a range of studies (e.g., Dejean et al. 2011; Mauvisseau et al. 2022; Schmidt et al. 2021) but are still difficult to control for. Similarly, shedding rates may vary strongly between taxa and can therefore affect detectability of eDNA (Dejean et al. 2011; Mauvisseau et al. 2022). Though needed for practical use, sampling times and volumes are nearly impossible to standardise and must be determined individually for each application. Nevertheless, comparative studies can help to develop broad guidelines for field protocols and enable practitioners to use eDNA metabarcoding (Bruce et al. 2021). Apart from these difficulties in sampling, laboratory protocols and bioinformatics may also considerably affect the results (Creedy et al. 2022; Hakimzadeh et al. 2023; Harper et al. 2019). Unlike conventional monitoring methods, eDNA results may contain both false‐negatives and false‐positives and contamination of samples can be an important source generating these errors (Ficetola, Taberlet, and Coissac 2016). However, false‐positive results can also occur during database matching, which is why strict quality control is required. Many studies reported hyper‐diverse communities through eDNA metabarcoding, implying a great, previously overseen, diversity, but reliable reference data, generated through alternative assessment methods, are often lacking (Fediajevaite et al. 2021).

A variety of studies exist using eDNA for amphibian assessments to track invasive (Dejean et al. 2012; Ficetola et al. 2008; Lin, Zhang, and Yao 2019) or endangered species (Rees et al. 2014; Takahara et al. 2020; Villacorta‐Rath et al. 2021), but scarcely to target whole community data. While the detection of invasive and endangered species has been successfully demonstrated through eDNA metabarcoding and quantitative polymerase chain reaction (qPCR) approaches, whole community assessments have so far shown mixed results in comparison to conventional methods (transect walks, dip netting and call‐counting) (Moss et al. 2022; Svenningsen, Pertoldi, and Bruhn 2022; Wikston et al. 2023).

For Odonata, single specimen DNA barcoding is a common tool for species identification, resulting in well‐covered reference databases (Galimberti et al. 2021; Geiger et al. 2021; Haring et al. 2020; Maggioni et al. 2021; Sittenthaler et al. 2023). However, to our knowledge, Odonata have not been investigated in detail through eDNA metabarcoding, although this comparatively small species group can be particularly suitable as indicator for biodiversity monitoring (Briers and Biggs 2003; D'Amico et al. 2004; Golfieri et al. 2016). Existing studies considered Odonata at the level of families or functional groups in broad datasets among other species groups (Sun et al. 2019; Zizka, Geiger, and Leese 2020) or with a focus on single species using qPCR instead of metabarcoding (Schmidt et al. 2021; Thomsen et al. 2012). Utilising eDNA metabarcoding from water samples to assess the Odonata community structure at the sampled waterbody has not been investigated yet.

Here, we assembled for the first time a comprehensive dataset on Odonata and amphibians based on elaborate field protocols, testing the suitability and accuracy of eDNA metabarcoding in comparison to conventional field sampling methods. We collected our data in a variety of ponds and lakes in extraction sites of building materials, which are of particular importance as habitats for these species groups (Bobrek 2020; Kettermann and Fartmann 2023). Our aim was to provide recommendations for suitable cost‐ and time‐efficient monitoring protocols for extraction companies that play an important role in the protection and promotion of the species under study.

We aim to answer the following questions: (i) Is eDNA metabarcoding suitable for amphibian and Odonata assessment in general? (ii) Can it provide better results than conventionally applied diversity assessment methods? (iii) Can independent use be recommended in terms of detection success of species, and time and money efficiency—or is the application limited to complement existing monitoring frameworks?

To answer these questions, we measured species numbers, cumulative species richness, community structure and detection probabilities assessed through eDNA metabarcoding and transect walks on N = 56 plots (six sites, 8–10 plots per site).

2. Materials and Methods

2.1. Study Design

Our study was part of the GiBBS project, a large multidisciplinary project focusing on biodiversity management in the building materials industry in Germany. We selected six extraction sites in Northwest Germany representative for the extracted geological substrates. Raw materials are extracted using dry (1 × gravel, 1 × limestone, 1 × sand) and wet (2 × gravel, 1 × sand) mining methods (Figure 1). Water bodies within extraction sites were selected based on satellite images, and their suitability for the assessment of amphibians and Odonata was evaluated by on‐site inspections. 8 to 10 representative transects were laid out per extraction site depending on the number of waterbodies available, resulting in a total of 56 plots. Depending on waterbody size, transects covered only a section of the littoral zone (larger ponds and lakes) or the whole waterbody (small, shallow ponds). The size of the water bodies was determined using aerial images with a minimum resolution of 1 m provided by Google Earth (Google LLC, Mountain View, CA, USA). For ponds with highly fluctuating water levels, the maximum size was used. Data and sample collection for all transects was conducted between 9th May and 11th August 2022. Transect walks for amphibian and Odonata assessment were repeated five times, while eDNA sampling was done twice during the second and fourth transect survey (Figure 1, details on data collection in the next section). Amphibian and eDNA sampling were conducted simultaneously, while Odonata were sampled with a medium time offset of 8 days. Some of the waterbodies dried out and could therefore not be included for further data collection. For detailed information on sampling procedures, see Appendix S2.

FIGURE 1.

Distribution of the sampling sites in Northwest Germany (left) and visualisation of the methods used, combining the data acquisition for amphibians, Odonata (via exuviae) and eDNA along defined transects (right).

2.2. Sample Collection and Processing

2.2.1. Amphibian Assessment

We conducted visual encounter surveys to assess amphibian diversity, abundance and community structure. Systematic transect walks were undertaken, following the same path for 30 min for all five time points, and amphibians were determined by sight and acoustics. In addition, every few steps (15 times in total), a dip net was pulled through the water and captured amphibians were identified and counted. For all detected amphibians, life stage (larvae, subadult [immature juveniles and yearlings] and adult) and sex were noted. The species of the genus Pelophylax ( P. esculentus , P. lessonae , P. ridibundus ) were not differentiated but treated as one species complex Pelophylax agg. due to close similarities because of hybridisation.

2.2.2. Odonata Sampling

The Odonata fauna was assessed using exuviae to obtain reliable evidence of reproduction (Raebel et al. 2010). For this purpose, the same transects as for the amphibians were systematically searched for 30 min and all exuviae found were collected for identification. In exceptional cases, collection of all exuviae was not possible within the specified time due to mass hatching events of individual damselflies. In these cases, as many exuviae as possible were collected along the entire transect length to cover all bank structures equally. Morphological identification of exuviae was carried out to species level when possible, using Brochard et al. (2016) and Gerken and Sternberg (1999). Where this could not be achieved due to damaged specimens or inconclusive identification characters, the result was given at the genus or family level. Because of the high similarity of their exuviae, Coenagrion puella and C. pulchellum were treated as a species complex.

2.2.3. Environmental DNA (eDNA) Sampling

Water samples for eDNA metabarcoding were collected two times per transect (06/07–06/16/2022 and 07/07–07/15/2022). Along the transect, three 500 mL subsamples were taken from the littoral zone in Nalgene plastic bottles. Water samples were pooled in a container and as much water as possible was filtered for 30 min through two different filter types (encapsuled 0.45 μm self‐preserving eDNA filter, polyethene sulfone membrane, Smith‐Root Inc., USA and 0.45 μm Thermo Scientific Nalgene filters, cellulose nitrate membrane, Thermo Fisher Scientific Inc., Waltham, MA, USA) using the eDNA Citizen Scientist sampler (Smith‐Root Inc.). Filtering time was determined to be 30 min to ensure comparable sampling effort to transects. For detailed information about filter volume, see Appendix S2. For each sample site, a blank filter was processed as a negative control. Self‐preserving filters were transported to the laboratory facilities and stored at −8°C until further processing, while Nalgene filters were transferred to 96% undenatured ethanol and stored at −20°C. A PCR amplicon‐free room was used for further processing of samples, and all used materials were cleaned with DNAway or 3% bleach and exposed to UV light for 20 min. Filters were cut in two halves, and half of the filter was further stored at −20°C. The other filter half was cut in small pieces, and the DNeasy 96 Blood and Tissue Kit (Qiagen) was used for DNA extraction following the manufacturer's instructions (to cover complete filter material, lysis volume was doubled to 400 μL). The kit implements sample filtration through a silica column and two washing steps to ensure sample purification and reduce amplification inhibition. Isolated DNA was eluted in 2 × 50 μL elution buffer. Per extraction plate, 12 negative controls (400 μL ATL lysis buffer [Qiagen]), which were processed separately through the whole laboratory protocol and sequencing, were added. Extraction success and negative controls were checked on a 1% agarose gel.

A two‐step PCR protocol was applied using standard Illumina Nextera primers (Oligonucleotide sequences 2019 Illumina Inc. All rights reserved.) for dual indexing of samples. To amplify amphibian DNA as well as other vertebrates, a 97 bp long fragment of the 12S gene was amplified using the primers 12SV05 forward (5′‐TTAGATACCCCACTATGC‐3′; Riaz et al. 2011) and 12SV05 reverse (5′‐TAGAACAGGCTCCTCTAG‐3′; Riaz et al. 2011). The first PCR (PCR 1) for marker gene amplification was performed in duplicates including 6.5 μL PCR Multiplex Plus Kit (Qiagen), 1 μL of DNA template and 0.2 μM of the forward and reverse primer. The PCR mix was filled up with 5 μL nuclease free water (ddH₂O) to a 13 μL reaction volume. The following PCR program was applied: initial denaturation at 95°C for 5 min; 30 cycles of: 30 s at 95°C, 30 s at 50°C and 50 s at 72°C; final extension of 5 min at 72°C. The PCR 1 product was used for the second PCR (PCR 2) for unique indexing of samples. Second PCR was also performed with the PCR Multiplex Plus Kit (Qiagen). Dual indexing of sequences (Nextera XT Index Kit v2 for library preparation) was applied to guarantee the assignment of sequences to the sample of origin in bioinformatic analysis. PCR 2 reaction included the 1 μL DNA template from PCR 1, 0.2 μM of each tagging primer (Nextera, Illumina, San Diego, CA, USA) and 12.5 μL master mix filled up with 9.5 μL H₂O. Primers included a nucleotide overhang as a binding site for primers in PCR 2, which was run with the following program: initial denaturation at 95°C for 5 min; 15 cycles of: 30 s at 95°C, 30 s at 50°C and 50 s at 72°C; final extension of 5 min at 72°C. PCR success was evaluated on a 1% agarose gel before PCR products were normalised using a SequalPrep Normalisation plate (Thermo Fisher Scientific, Walton, MA, USA) following the manufacturer's instructions with an end concentration of 25 ng per sample (100 μL). For each sample, 10 μL was pooled together, and left‐sided size selection with magnetic beads was applied twice on the sample pool to remove primer residuals (ratio 0.85x, SPRIselect Beckman Coulter). Library concentration was measured with a Quantus fluorometer (Promega, Madison, WI, USA) and on a FragmentAnalyzer (Agilent Technologies, Santa Clara, CA, USA), and the pool was sent for sequencing on one Novaseq 6000 S4 (2 × 150 bp) to Macrogen Europe B.V., Netherlands.

For Odonata DNA amplification and other invertebrates, a 313 bp long fragment of the COI gene was amplified using the primers fwhF2 forward (5′ GGDACWGGWTGAACWGTWTAYCCHCC‐3′, Vamos, Elbrecht, and Leese 2017) and Fol_degen_rev reverse (5′‐TANACYTCNGGRTGNCCRAARAAYCA‐3′, Yu et al. 2012). This highly degenerated primer pair is mainly used for insect bulk sample amplification and targets a wide range of aquatic, terrestrial and amphibiotic insect orders (Elbrecht et al. 2019). Before laboratory processing, primers were mapped in silico against available mitogenomes of German Odonata species (mitogenomes available for 45 out of 82 species, covering all prevailing families). No critical mismatches (first three bases 3′ end) were identified for any of the compared sequences indicating better performance in Odonata amplification than the primers mICOIintF/dgHCO2198 (Leray et al. 2013) and EPTD2nr (Leese et al. 2021).

The first PCR (PCR 1) for marker gene amplification was performed in triplicates with same volumes as for Amphibia listed above. The following PCR program was applied: initial denaturation at 95°C for 5 min; 30 cycles of: 30 s at 95°C, 30 s at 50°C and 50 s at 72°C; final extension of 5 min at 72°C. The product of PCR 1 was used for the second PCR (PCR 2) for unique indexing of samples. PCR 2 was performed with the same volumes as for Amphibia and the same PCR protocol as in PCR 1 but with 15 cycles. PCR success was evaluated on a 1% agarose gel before PCR products were normalised using a SequalPrep Normalisation plate (Thermo Fisher Scientific, MA, USA) following the manufacturer's instructions with an end concentration of 25 ng per sample (100 μL). For each sample, 10 μL was pooled together, and left‐sided size selection with magnetic beads was applied twice on the sample pool to remove primer residuals (ratio 0.7×, SPRIselect Beckman Coulter). Library concentration was measured with a Quantus fluorometer (Promega) and on a FragmentAnalyzer (Agilent Technologies, Santa Clara, CA, USA) and the pool was sent for sequencing on one Hiseq 2500 run (2 × 250 bp) to Macrogen Europe B.V., Netherlands.

2.3. Data Analysis

2.3.1. eDNA Metabarcoding

Raw sequences were analysed as follows: PCR primers were removed using cutadapt4.4 (Martin 2011), and paired‐end merging was conducted using vsearch2.15 (‐fastq_maxdiffs 5, ‐fastq_pctid 90) (Rognes et al. 2016). Quality filtering (‐fastq_maxee 0.5) and frequency and length filtering (‐minsize 2, ‐fastq_minlen fragment length−10, ‐fastq_maxlen fragment length + 10) was conducted with vsearch2.15 as well as extraction of ASV (Amplicon Sequence Variants) using ‐cluster_unoise ‐minsize 4 ‐unoise_alpha 2. Further filtering for nuclear pseudogenes was performed with the metaMATE (Andújar et al. 2021), followed by de novo chimera filtering (‐uchime3_denovo, vsearch2.15). Operational Taxonomic Units (OTUs) were clustered based on a 97% similarity threshold, and reads per OTU were mapped into an OTU table. Read numbers present in negative controls were subtracted from regarding samples (12 filter negative controls per dataset and 42 extraction negative controls for Odonata dataset, 28 extraction negative controls for amphibian dataset), and replicates per sample were merged. Singletons per sample were removed. Taxonomic assignment of OTUs was conducted by comparison with a custom reference database for Arthropoda (COI) and amphibians (12S). The database was created using taxalogue (Noll, Scherber, and Schäffler 2023) with sequences and taxonomies obtained from BOLD Systems (BOLD) (Ratnasingham and Hebert 2007), GenBank (NCBI) (Sayers et al. 2022) and the German Barcode of Life (GBOL) (German Barcode of Life Consortium 2023; Geiger et al. 2016) accessed on 14th of July 2023. Taxon names were normalised according to the NCBI Taxonomy (https://www.ncbi.nlm.nih.gov/taxonomy). A gap analysis for all species found by transect walks revealed the availability of reference sequences for all species. Taxonomic assignment was conducted using the SINTAX classifier integrated in vsearch2.15 with 85% cutoff (vsearch ‐sintax molecular.units.fas ‐db laxalogue.database.fas ‐sintax_cutoff 0.85 ‐output.tsv). Only OTUs with species level assignment were included for downstream analysis, and if several OTUs were assessed per taxonomic assignment, those were merged and read numbers were added up.

2.3.2. Statistical Analysis

The statistical analysis was performed in R Version 4.3.1 and RStudio Version 2023.6.1.524 (R Core Team 2023; Posit team 2023). For both molecular and conventional data, only cases where determination to the species level was achieved were included in the analyses. For the conventional amphibian data, all occurrences independent of life stages were counted as species evidence. For the molecular amphibian data, results of two markers (12S and COI) were available but based on the Euler diagram (Figure 2), only the 12S results were used for further comparison with the conventional data.

FIGURE 2.

Euler diagrams of all species assessed by eDNA vs. transect walks. For amphibians, the results of two marker genes (COI and 12S) are shown. None of the 11 species detected were found exclusively by COI or transect walks. At the same time, none of the 29 Odonata species were found exclusively by eDNA. The figure only includes species level assignments; for Odonata, further molecular units were assigned to the genus levels Anax and Coenagrion.

As primary data exploration showed small differences in diversity patterns between the complete and the reduced dataset (including only the two sampling events for transect walks congruent with eDNA sampling) (Figure 1), results are shown for the complete data, while all results for the reduced set are included in Appendix S9.

Differences in species numbers between sampling methods were tested using Wilcoxon signed rank tests. To analyse the influence of sampling effort, rarefaction curves were calculated using the iNext package (Chao et al. 2014; Hsieh, Ma, and Chao 2016). Community distance matrices were calculated with the vegdist function (package vegan [Oksanen et al. 2022]) to test whether the assessed species communities differed between the methods. The PERMANOVAs used for this were calculated with the adonis2 function, which is included in the vegan package as well.

Occupancy models (MacKenzie et al. 2002) were calculated with the unmarked package (Fiske and Chandler 2011; Kellner et al. 2023) for each species detected by both methods to account for false negative detection and to compare detection probabilities for individual species between the applied methods. Therefore, detection matrices for each species were generated including presence/absence data for the 56 plots and all sampling events. The sampling method was used as explanatory variable with the levels ‘eDNA’ and ‘transect walks’. Predictions of detection probabilities including standard errors were made with the appropriate predict method for objects of class ‘unmarked’. Details on model outputs and predictions can be found in Appendix S5.

3. Results

Overall, we found 11 amphibian species, of which eight could be detected by both eDNA metabarcoding and transect walks (Figure 2). None of the 11 species were found exclusively by transect walks or COI marker amplification. As COI is not the standard marker for vertebrate metabarcoding assessments, these results will not be considered further here. Two species, Rana dalmatina and Lissotriton helveticus , were detected exclusively by eDNA amplification. These records overlapped with the natural range of these species that show a restricted and scattered distribution in Central Europe. A complete list of species including frequencies can be found in Appendix S1.

Out of the 29 Odonata species, 10 were detected by both methods (Figure 2). The genera Anax and Coenagrion were also found by eDNA, but as database comparison did not reach species level assignments, they could not be included. In contrast to the amphibians, there were no species detected exclusively by eDNA, resulting in 19 species being missed by this method. A comparison of the exuviae numbers of the species found and not found through eDNA metabarcoding showed that the detected species were significantly more abundant (p‐value = 7.443e−14, Appendix S3). Among the detected species, Anisoptera (dragonflies) and Zygoptera (damselflies) were equally represented.

The species numbers per plot showed different patterns for the two groups. Amphibian species numbers ranged from 0 to 5 and did not differ between the methods (p‐value = 0.71), while Odonata species numbers ranged from 0 to 10 and were significantly higher for the transect walks (p‐value = 2.7e−7, Figure 3a). This was also reflected in the mean proportions of species detected by the respective methods per plot (Figure 3b). For the amphibians, eDNA detected 65% and transect walks 75% of the total species, resulting in an average overlap of 40% of species found by both methods. Of the total number of Odonata, however, 30% of the species were detected by eDNA, while 90% were detected by transect walks, resulting in a mean overlap of 20% between the two methods. Hence, the methods led to more similar results for the amphibians than for the Odonata. Nevertheless, the two methods assessed communities that differed significantly for both the Odonata and the amphibians, as shown by the PERMANOVA results (Odonata p‐value = 0.001, amphibians p‐value = 0.001, Appendix S4).

FIGURE 3.

(a) Mean species numbers per plot for each survey method and species group. Differences in group means were tested by a Wilcoxon test (Amphibians p = 0.71, Odonata p = 2.7e−7). (b) Detection quality, expressed as mean proportions of detected species per plot, including the intersection between the two methods. (c) Species accumulation curves for the number of samples within both survey methods (eDNA n = 106, transect walks n = 267) and species groups.

In terms of sampling effort, eDNA recorded a higher number of amphibian species with fewer samples compared to the transect walks (Figure 3c). The species richness of the transect walks almost reached saturation with the given number of samples but remained below the level of the eDNA method. The corresponding results for the Odonata were fundamentally different. The species richness of the transect walks increased strongly with the number of samples and finally reached a saturation level. For the eDNA assessment, however, species richness increased only slightly and remained well below the transect walks at a maximum of 10 species.

The comparison of the detection probabilities for all species assessed by both methods showed mixed results (Figure 4; Appendix S5). Among amphibians, Pelophylax agg. and Lissotriton vulgaris had the highest detection probabilities, which were similar between methods for Pelophylax agg. and significantly higher with eDNA for L. vulgaris . Triturus cristatus and Ichthyosaura alpestris were also more likely to be detected by eDNA, while Rana temporaria was the only amphibian for which a higher detection probability through transect walks was found. Among Odonata, all significant differences in detection probabilities were in favour of the transect walks (Ischnura elegans, Enallagma cyathigerum and Erythromma lindenii). Due to small sample sizes, large standard errors occurred for some model predictions.

FIGURE 4.

Detection probabilities depending on the survey method. All amphibians (a) and Odonata (b) detected by both methods are shown. Filled circles indicate significant difference between the methods. The predictions of detection probabilities are based on single species occupancy models using the sampling method as an explanatory variable (see Appendix S5 for model coefficients and estimates).

Although it might have been expected, the different number of survey replicates (5 for the transects and 2 for the eDNA) was not the cause of differences between the methods. Performing all our analyses with the reduced data set (including only the two surveys congruent for both methods) did not result in any change in the patterns of total species list, species numbers and detection probabilities (Appendix S9). Only the detection rates of amphibians showed a shift towards better detection by eDNA. Thus, we can exclude that the different number of survey replicates (5 for the transects and 2 for the eDNA) was the cause of the differences between methods.

4. Discussion

In the present study, we collated a dataset of 56 sampling points in mining sites of the building materials industry to assess the suitability of eDNA metabarcoding for amphibian and Odonata community monitoring. Overall, we found strong overlaps between molecular and conventional methods for amphibians, but major differences for Odonata.

Although species‐specific eDNA approaches such as qPCR can yield reliable information on known species assemblages reducing problems like species masking or primer bias (Bylemans et al. 2019; Harper et al. 2018; Schneider et al. 2016; Wood et al. 2019), our metabarcoding approach proves more useful for cases where the target species are unknown. Thus, for a comprehensive survey of aquatic biodiversity, qPCR approaches would not have been possible in the first place. Metabarcoding has been shown to be cheaper for large numbers of samples and target species (Hänfling et al. 2017; Harper et al. 2018; Wilcox et al. 2020), providing greater certainty and reproducibility through generation of DNA sequences. Our study thus contributes to testing and improving eDNA metabarcoding, especially for small waterbodies, for which only few data exist now.

4.1. Amphibians

For amphibians, eDNA metabarcoding yielded results that were as good as or better than conventional methods, depending on the comparison criteria. Both methods demonstrated good detection quality and showed a significant overlap of the detected species. However, each method also provided several unique observations, indicating that they also complement each other. Especially for newts, detection probabilities were higher for eDNA metabarcoding than for transect walks due to the group's close association with the waterbody during summertime and a presumably constant DNA release. Together with the higher detection probability derived from transect walks for less water‐bound species such as Rana temporaria , this illustrates a strong dependency of metabarcoding detection sensitivity on species‐specific behavioural patterns (Dunn et al. 2017; Barnes and Turner 2016). Thus, a combination of methods to achieve high‐quality biodiversity estimates is likely most promising. This is further supported by our finding that, although there is high similarity in alpha‐diversity, the methods showed significant differences at the community level. Previous studies have shown that combining multiple methods may yield more complete monitoring data (Moss et al. 2022; Wikston et al. 2023). Although this would be the most desirable approach to arrive at highest possible data quality, it is often impractical because of high sampling effort. Even implementing just one single, standardised monitoring method can be difficult for reasons of time and costs in practical nature conservation. From a practitioner's perspective, our results show that just two eDNA samplings, instead of multiple transect walks, can be sufficient for amphibian sampling. A minor caveat is that this approach only yields presence/absence data. Nevertheless, where simplicity is crucial and obligatory to realise monitoring, two runs of eDNA sampling deliver reliable data if other prerequisites such as suitable sampling time and lab protocols are met. Although our surveys had been done comparatively late in the season, we were able to detect a wide spectrum of potential species. However, a loss of signals after metamorphosis, such as reported by Brys et al. 2021, was also visible in our data for Rana arvalis , R. temporaria and Bufotes viridis as those species were not detected in the second eDNA survey anymore. Thus, adjusting sampling time could still increase the number of species detections.

Our lab and sampling protocol enabled reliable assessments of amphibian diversity and can be recommended for practical mapping exercises. Even using CO1 primers yields a well‐covered spectrum of species comparable to transect walks—with the additional benefit of also giving insights into aquatic invertebrates. Unfortunately, it was not possible to reliably distinguish between the morphologically similar Anuran species Rana lessonae, Rana ridibunda and the hybrid Rana kl. ‘esculenta’ in our eDNA results. Based on 12S identification, 100% similarity to deposited sequences of all three species was observed. We therefore summarised detections in the complex Pelophylax agg. For future studies, using a second marker gene might improve resolution within this complex.

4.2. Odonata

We found significant differences between eDNA metabarcoding and exuviae assessment for Odonata at the species and community level, in total indicating that eDNA metabarcoding, as applied here, cannot deliver reliable data on Odonata. Although the detection probabilities of the species actually found by eDNA were similar to those of the transect walks, particularly rare species with low abundances could not be detected at all and must be taken into account more specifically.

The amount of DNA in the waterbody appeared to be crucial, as the species detected by metabarcoding were significantly more abundant in terms of exuviae numbers. This is in line with previous studies, showing that species detection through eDNA strongly depends on biomass and density of target organisms (Dejean et al. 2011; Jo et al. 2019; Schmidt et al. 2021). Consideration should be given to the idea that Odonata are not necessarily ideal for water sample eDNA detection because of their low biomass and seasonality. We would therefore like to encourage more studies, such as the one on crayfish by Dunn et al. (2017) to improve knowledge on correlations between life cycle and amounts of eDNA for Odonata. Alternatively, taking soil samples could also provide interesting insights into the ecology of dragonfly eDNA, as outlined by Barnes and Turner (2016). Either way, adjustments in sampling design and replication effort in the field could improve metabarcoding results with water samples. Sampling should therefore be timed to coincide with larval presence in the water body to maximise the probability of capturing target DNA. In our case, this prerequisite was met, as we were able to prove species' presence during eDNA sampling through our ‘ground‐truthing’ exuviae data (Appendix S7). The time offset between eDNA and conventional sampling had no influence on this, as there was no correlation with method similarity (Appendix S6). Since the Odonata larvae have residence times of weeks to months in the water, we assume that the target DNA was present during our sampling.

Nevertheless, many species were missed by eDNA, which implies the presence of other factors preventing detection. As taxa with solid exoskeletons, such as Odonata larvae, release only small amounts of DNA into the water (Schmidt et al. 2021), an increase in filtered water volume or spatial replicates may be considered as solutions enhancing capturing of target DNA. However, we did not find any correlation between filter volume and method similarity in our data (Appendix S6).

Although filtering and analysing each spatial replicate separately instead of pooling could have yielded to a higher number of detected species, this is not advisable from a practical point of view. Further, our data suggest that an increase in temporal replicates (more site visits) would not have improved eDNA data. All potential species were present during the two sampling events, and effort should therefore better be spent on increased spatial replicates, where possible. This is different from what Moss et al. (2022) found in their study on the detection sensitivity of eDNA metabarcoding for amphibians as their results rather implicated expanding visits than filter volume or replication of other protocol steps.

The replication during the laboratory analyses (filtration, extraction and PCR) has been shown to increase the probability of detecting species in general and rare taxa in particular (Bruce et al. 2021; Ficetola et al. 2015; Hestetun, Lanzén, and Dahlgren 2021; Murray, Coghlan, and Bunce 2015). Our PCR replicate number was limited to what could be realised in terms of time and money. Many Odonata species were only detected in one of the six PCR replicates (Appendix S8), showing that high replicate numbers will be needed to reliably detect all species, including rare ones with low biomass (Dickie et al. 2018; Ficetola et al. 2015; Hestetun, Lanzén, and Dahlgren 2021).

In addition to determining appropriate replication effort in the laboratory analyses, suitable marker genes and primers are required. COI is the commonly used marker for eDNA metabarcoding of invertebrates. The degenerated primer we used amplifies a broad taxonomic range of invertebrates but at the same time shows high ‘untargeted amplification’ such as bacteria, protists or algae. The amplification of untargeted taxa reduced available reads for Odonata, potentially leading to undetected species with low DNA concentration in the sample (Macher et al. 2018; Zizka, Geiger, and Leese 2020). An adapted primer pair attempting to perform a more specific DNA amplification of aquatic macroinvertebrates (EPTD2nr) (Leese et al. 2021) was also considered but finally not used due to mismatches for Odonata species in in silico evaluation. In addition, single‐specimen barcoding of Odonata revealed that the species Anax spp. ( A. imperator , A. parthenope) and Coenagrion spp. ( C. puella , C. pulchellum and C. ornatum ) show shallow genetic distances and sporadic haplotype sharing (Galimberti et al. 2021; Geiger et al. 2021) and are therefore problematic to distinguish based on the COI marker gene sequences (Sittenthaler et al. 2023). The Cytb and ND1 (NADH dehydrogenase I) gene regions have been proposed as additional markers for Odonata identification, which should therefore be considered together with more specific primers to improve taxonomic resolution in future studies.

Finally, the taxonomic assignment of the OTUs through database comparison was not influenced by gaps in the reference data, as none were available. Moreover, using the SINTAX approach ensured a strict comparison with the reference libraries so that assignments were highly reliable and reached species level for the majority of OTUs. While previous studies found no interspecific delimitation between A. parthenope and A. imperator , a reanalysis of our data based on Amplicon Sequence Variants (ASVs) (instead of OTUs delimited at 3% threshold) allowed us to clearly separate the haplotypes to the two species. In contrast, the ASV‐based reanalysis did not lead to distinct assignments of C. puella and C. pulchellum , most probably because of a lack of genetic variability or inconsistencies in reference sequences due to morphological similarity of species (Noll, Scherber, and Schäffler 2023).

Since we used the results of the exuviae surveys as a basis for comparison, it should be mentioned that these are not unbiased either (Bried, D'Amico, and Samways 2012; Giugliano, Hardersen, and Santini 2012; Raebel et al. 2010) and an extension of the method comparison between DNA and multiple conventional methods (such as in Moss et al. 2022 and Wikston et al. 2023 for amphibians) would possibly provide further insights in method suitability and detection probabilities. Regarding Odonata as a target species group for eDNA metabarcoding, we can conclude that testing other markers and primers and increasing replication will be necessary to improve detection sensitivity.

5. Conclusion

Both traditional and molecular methods were suitable and comparable for amphibian assessment. Metabarcoding detected species more efficiently and should therefore be considered in future monitoring with reduced sampling effort compared to conventional assessment. We found surprisingly good results for amphibians with the use of COI. The species spectrum was largely covered, but reads and frequencies were considerably lower than for the 12S results.

For Odonata, our study uncovered considerable gaps in the metabarcoding results. We recommend evaluation and improvement in community assessment based on eDNA, considering the design of more specific primers expanding into other gene regions than COI and the application in multiplex PCR approaches. Additionally, replicating several protocol steps may improve detection success, though it requires more time and costs, which could hinder its application in monitoring efforts. If target species are clear and time and costs are not limiting factors, a species‐specific approach using qPCR should be considered to obtain reliable results for Odonata from eDNA samples. Overall, our study contributes to establishing eDNA metabarcoding for amphibians as an equivalent method alongside conventional surveys in biodiversity monitoring and lays the foundation for a reliable assessment of Odonata using this method.

Author Contributions

K.S. contributed to designing the research, performing conventional Odonata assessments, analysing the data and writing the manuscript. V.Z. contributed to designing the research, performing eDNA sampling and sequencing, analysing the data and writing the manuscript. C.S. contributed to designing the research, analysing the data and editing the manuscript. N.H. contributed to designing the research and editing the manuscript. All authors gave their approval for the final version of the manuscript.

Conflicts of Interest

The authors declare no conflicts of interest.

Benefit‐Sharing Statement

Benefits from this research accrue from the sharing of our data and results on public databases as described above.

Supporting information

Appendices S1–S9.

Data Availability Statement

Raw data can be found at the European Nucleotide Archive (Accession number PRJEB73319). Raw OTU tables can be found on figshare (https://doi.org/10.6084/m9.figshare.26067520.v1). Further data and scripts are openly available at Dryad (DOI: 10.5061/dryad.v41ns1s61).