Layout 1 INTRODUCTION Since its first application, the use of environmental DNA (eDNA) isolated from water samples to detect the presence of taxa has been considered a promising method to improve aquatic biomonitoring (Ficetola et al., 2008; Goldberg et al., 2015; Lawson Handley, 2015). Environmental DNA is the genetic material pres- ent in different environmental matrices such as sediment, water, and air, and belonging to the organisms inhabiting the surveyed area; it includes DNA released in the envi- ronment (intra or extracellular) and DNA taken directly from living cells (Pawlowski et al., 2018). This eDNA can be extracted from the environmental matrices and used to assess community biodiversity through the am- plification of a short DNA region used as a “barcode” (Hebert et al., 2003; Taberlet et al., 2012; Ward et al., 2009) and sequenced with high-throughput sequencing techniques (i.e., metabarcoding, Kuntke et al., 2020; Schenekar et al., 2020). The use of eDNA metabarcod- ing can significantly improve biodiversity monitoring surveys through the early detection of exotic and poten- tially invasive species and the tracking of elusive endan- gered species (Deiner et al., 2018; Pawlowski et al., 2018; Taberlet et al., 2018). A range of organisms is used worldwide as indicators (Biological Quality Elements (BQEs)) to monitor the qual- ity status of aquatic ecosystems, namely phytoplankton, phytobenthos, aquatic plants, macroinvertebrates, and fish (European Environment Agency, 2018). The sampling method officially recommended for the biodiversity assess- ment of river and lake fish by the Water Framework Direc- tive (EN14011, 2003) are electrofishing and gill netting, which are quite expensive methods and require a large and qualified staff to be performed. Moreover, several sampling practices (e.g., seines or trawling) can severely damage the habitat and in large lakes can heavily affect fish abundance (Irvine et al., 2019; Njiru et al., 2018). Since 2000, the European Union has been actively engaged in the protection and enhancement of aquatic ecosystems: freshwater biomonitoring promoted by the implementation of the EU Water Framework Directive (WFD) (European Commission, 2000) allows for the evaluation and improvement of their quality status. However, classical biomonitoring practices require good taxonomic expertise and the probability of detect- ing species that represent <1% of the total abundance is ARTICLE Alpine freshwater fish biodiversity assessment: an inter-calibration test for metabarcoding method set up Giulia Riccioni1*, Isabelle Domaizon2, Andrea Gandolfi1, Massimo Pindo1, Adriano Boscaini1, Marine Vautier2, Hans Rund3, Peter Hufnagl4, Stefanie Dobrovolny4, Valentin Vasselon5, Jonas Bylemans1,6, Cuong Q. Tang7, Nico Salmaso1, Josef Wanzenböck3 1Research and Innovation Centre, Fondazione Edmund Mach, S. Michele all'Adige, Italy; 2National Institute of Research for Agriculture, Alimentation and Environment, University Savoie Mont Blanc, Pole R&D ECLA, Thonon-les-Bains, France; 3University of Innsbruck, Research Department for Limnology, Mondsee, Austria; 4Austrian Agency for Health and Food Safety GmbH, Wien, Austria; 5French Office for Biodiversity, Pole R&D ECLA, Thonon-les-Bains, France; 6University of Lausanne, Department of Ecology and Evolution, Biophore, Lausanne, Switzerland; 7NatureMetrics Ltd., Egham, UK ABSTRACT The analysis of environmental DNA (eDNA) by high throughput sequencing (HTS) is proving to be a promising tool for fresh- water fish biodiversity assessment in Europe within the Water Framework Directive (WFD, 2000/60/EC), especially for large rivers and lakes where current fish monitoring techniques have known shortcomings. These new biomonitoring methods based on eDNA show several advantages compared to classical morphological methods. The sampling procedures are easier and cheaper and eDNA metabarcoding is non-invasive and very sensitive, allowing for the detection of traces of DNA. However, eDNA metabarcoding methods need careful standardization to make the results of different surveys comparable. The aim of the EU project Eco-AlpsWater is to test and validate molecular biodiversity monitoring tools for aquatic ecosystems (i.e., eDNA metabarcoding) to improve the traditional WFD monitoring approaches in Alpine waterbodies. To this end, an inter-calibration test was performed using fish mock community samples containing either tissue-extracted DNA, eDNA collected from aquaculture tanks and eDNA samples collected from Lake Bourget (France). Samples were analysed using a DNA metabarcoding approach, relying on the amplification and HTS of a 12S rDNA marker, in two separate laboratories, to evaluate if different laboratory and bioinformatic protocols can provide a reliable and comparable description of the fish communities in both mock and natural samples. Our results highlight good replic- ability of the molecular laboratory protocols for HTS and good amplification success of selected primers, providing essential in- formation concerning the taxonomic resolution of the 12S mitochondrial marker in describing the Alpine fish communities. Interestingly, different concentrations of species DNA in the mock samples were well represented by the relative DNA reads abun- dance. These tests confirm the reproducibility of eDNA metabarcoding analyses for the biomonitoring of freshwater fish inhabiting Alpine and peri-Alpine lakes and rivers. No n- co mm er cia l u se on ly G. Riccioni et al.12 very low (Paller, 1995), thus providing partial estimates for entire communities (Deiner et al., 2017). Compared to electrofishing and gill netting procedures, water sam- pling procedures for eDNA analyses proved to be po- tentially cheaper and easier, non-invasive, and suited for surveys in extremely difficult sites. Reduced opera- tional costs could allow for regular sampling during the year, providing time-series data and a systematic mon- itoring of fish biodiversity (and, in general, of commu- nity diversity) in different seasons and extreme events (extremely dry seasons or floods). Conversely, tradi- tional sampling can usually be performed once or twice a year. Standardization of all the protocols, from the sam- pling activities to the taxonomic assignment of DNA se- quences (Dickie et al., 2018), is paramount to allow for the comparability among eDNA metabarcoding studies for the ecological assessment of habitats or ecosystems (Goldberg et al., 2015) and to improve the sensitivity of metabarcoding assays. The use of experimental controls and mock communities allows to exclude unspecific sig- nals and verify the recovery of species signals and quan- titative representation of species in the samples, as well as to determine bioinformatic filtering steps, and thresh- old levels. Using standardized methods allows for a bet- ter interpretation of ecosystem response to pressures (Mock and Kirkham, 2012; Morales and Holben, 2011) however, each study has its specificity and often requires a customization of the metabarcoding protocols that often need further validation, especially when universal polymerase chain reaction (PCR) primers are used to ex- plore fish community diversity (Bylemans et al., 2018; Thalinger et al., 2021a). Pilot studies, comprising inter-calibration tests, are an invaluable tool to evaluate if different approaches can lead to similar and comparable results (Zinger et al., 2019). It is for these reasons that, in this study, we per- formed an inter-calibration exercise involving two dif- ferent laboratories to evaluate the reproducibility and possible limits of an eDNA metabarcoding procedure to describe the fish community diversity in Alpine lakes and rivers. No previous studies were performed to esti- mate Alpine fish biodiversity by using eDNA metabar- coding, and a pilot test was highly recommended. Mock samples made of tissue DNA pools as well as aquaria water samples and lake water samples were included in the test to compare the species detection performance and replicability of the method. This test has been con- ceived within the Eco-AlpsWater network, a project funded by the European Union, whose ultimate goal is to evaluate and validate emerging technologies based on eDNA metabarcoding for the biodiversity assessment of freshwater ecosystems in the Alpine region (https://www.alpine-space.eu/projects/eco-alpswater). MATERIALS Set-up of mock and environmental samples All the samples were prepared at INRAE (Institut Na- tional de Recherche pour l’Agriculture, l’Alimentation et l’Environnement) as a blind test for the laboratories in- volved in the inter-calibration exercise. A negative control sample (pure water filtered following the same protocol used for the other samples) was included in the test and analysed following the same procedures (DNA extraction, library preparation, and sequencing) used for the other samples. Three mock samples including an increasing number of species pooled (6, 9, and 14, hereafter M6, M9, and M14) and different DNA proportions (Table S1) were set up using genomic DNA extracts from fin clips. We used one individual per fish species to set up these three mock samples. The fish species used to assemble these mock samples are species commonly found in lakes and rivers of the Alpine and perialpine regions. A fourth mock sam- ple was prepared by collecting 100 mL of water from each of ten fish tanks containing one single species each (for a total of seven species, namely Salmo carpio L., On- corhynchus mykiss W., Barbus caninus B., Perca fluvi- atilis L., Lepomis gibbosus L., Carassius carassius L., and Tinca tinca L.). For each tank, we collected the same volume of water to obtain a total volume of 1 L to simu- late a real environmental sampling procedure. This sample was collected on the 3rd of December, 2019, and extracted on the 9th of December, 2019. Moreover, a third set of three environmental samples of 6 x 2 L each were col- lected from three different areas in Lake Bourget (all sub- surface samples, i.e., 10-20 cm; Figure 1). Three sampling points were sampled from each lake bank to simulate a transect and 2 L of water were collected from each site (Figure 1). After careful mixing of the samples, 1 L of water collected from the fish tank sample and Lake Bour- get samples were filtered by using 0.45 µm SterivexTM capsule filters (Merck Millipore, Burlington, USA); the filtration cartridges were filled with SPYGEN (SPYGEN, Le Bourget du Lac Cedex, France) preservation buffer and stored at room temperature until the DNA extraction step. These lake samples were collected on the 16th of October, 2019, stored at room temperature, and extracted on the 21st and 22nd of October, 2019. DNA extraction and library preparation For fin clips DNA extractions, we used the Nucle- oSpin® DNA RapidLyse kit from MACHEREY-NAGEL. eDNA extraction was performed using the NucleoSpin® Soil kit (MACHEREY-NAGEL, Allentown, USA) follow- ing the protocol described in Pont et al. (2018) and adapted to SterivexTM capsule filters. Twenty µL of genomic DNA No n- co mm er cia l u se on ly eDNA metabarcoding test for fish biodiversity 13 extract were delivered to both Fondazione Edmund Mach, Italy (LAB_A) and NatureMetrics, UK (LAB_B) sequenc- ing platforms for further laboratory processing. For both laboratories, a hypervariable region of 12S rRNA was amplified via a two-step PCR process (Sup- plementary Information). The two laboratories performed all PCRs in the pres- ence of both a negative and a positive control (i.e., a mock community with a known composition of fish species). Amplification success at each step was determined by gel electrophoresis. All PCRs replicates per sample were pooled and purified using CleanNGS beads (CleanNA, Waddinxveen, Netherlands) by LAB_A and MagBind To- talPure NGS (Omega Biotek, Norcross, USA) magnetic beads with a ratio of 0.8:1 (beads: DNA) by LAB_B, to remove primer dimers. The sequences, saved in FASTQ formatted files, were deposited to the European Nucleotide Archive (ENA) with study accession number PRJEB49223. Bioinformatic analyses Bioinformatic analyses were performed by Lab_A using the OBITools3 software (Boyer et al., 2016). A ref- erence database was created simulating a PCR amplifica- tion in silico by using the ecoPCR program: the whole vertebrate EMBL database (June 2020) and the MiFish- U primers were used, allowing for a maximum of three mismatches with the published sequences, and blocking the last two nucleotides in the primer sequence. For FASTQ Miseq sequences analyses the alignpairedend script was used to perform a micro-assembly of paired- end reads and sequences with Illumina FASTQ quality scores <30 across the head, tail, or total length of the se- quence were discarded. The ngsfilter script was used to assign the reads to each sample through barcode identifi- cation and, after a dereplication step, only sequences longer than 80 nucleotides and a count ≥10 were retained for further analyses. The obiclean script was used to detect the potential PCR errors, selecting only sequences with the ‘head’ status and abundance higher than 0.05%. The taxonomic assignment was performed by using the ecotag script and the reference database, considering a 97% of similarity. The taxonomic assignment was further in- spected by using BLASTn (Zhang et al., 2000) algorithm optimized for very similar sequences (megablast) on the nucleotide collection (nr/nt) that includes all GenBank + EMBL + DDBJ + PDB sequences when uncertainties in the identification emerged. Rarefaction analysis was per- formed on Lake Bourget sequences to evaluate if the se- quencing effort allowed to reach a plateau using the vegan package (Oksanen, 2016) and the rarecurve function in R environment (R Core Team, 2020). To evaluate the intra-individual variability of the 12S rDNA copies, both Salmo trutta Miseq sequences as- signed at the species level and those assigned only at the genus level and obtained from the mock tissue sample (DNA extracted from the fin clip of a single individual), were further evaluated by using MEGAX (Kumar et al., 2018). These sequences were aligned separately by using the Muscle software (Edgar, 2004) with default parame- ters, and the mean distance among sequences was com- puted by using the Kimura 2-parameter model (K2P, Kimura, 1980) and 1000 bootstrap. At LAB_B samples were demultiplexed based on the combination of the i5 and i7 index tags. Paired-end reads for each sample were merged with USEARCH (Edgar, 2010), with a minimum overlap of 20% of the total read length. Forward and reverse primers were trimmed from the merged sequences using cutadapt (Martin, 2011), and retained if the trimmed length was between 140 and 200 bp. These sequences were quality filtered with USE- ARCH to retain only those with an expected error rate per base of 0.05 or below, and dereplicated by sample, retain- ing singletons. Unique reads from all samples were de- noised in a single analysis with UNOISE (Edgar and Flyvbjerg, 2015), requiring retained ZOTU`s (zero-radius OTU’s) to have a minimum abundance of 8 in at least one sample. A taxon-by-sample table was generated by map- ping all dereplicated reads for each sample to the ZOTU representative sequences with USEARCH, at an identity threshold of 97%. ZOTU’s were identified via BLASTn (Zhang et al., 2000) searches of the representative se- quences against the whole nt database and a local curated database of 12S fish sequences. Identifications were based Figure 1. Sampling sites map of Lake Bourget. Three 6 x 2 L samples were collected in three different areas of the lake and 1 L of water was filtered to evaluate the effectiveness of the eDNA metabarcoding survey in describing the fish community diver- sity (BouB = low Lake Bourget, BouM = medium Lake Bourget, BouH = high Lake Bourget). No n- co mm er cia l u se on ly G. Riccioni et al.14 on the highest available percentage identity at 98–100%, with an e-score of 1e-20 and a hit length of at least 80% of the query sequence. In cases where multiple reference se- quences matched equally to the query sequence then a more conservative higher taxonomic classification was considered. Only sequences with species- or genus-level identifications were included in the final results. When a species was represented by multiple ZOTUs, the one with the highest percentage match to that species was taken as the representative. Typically, the other sequences having the same occurrence pattern and the lower sequence sim- ilarity can be attributed to PCR or sequencing errors. Data analysis Analysis of covariance (ANCOVA) was used to com- pare two regression lines by testing the effect of a cate- gorical factor (the two laboratories involved in the high throughput sequencing (HTS) analyses) on the dependent variable (fraction of taxonomic annotated sequences) while controlling for the effect of the continuous covari- able (fraction of DNA of single species in mock tissue samples; Crawley, 2005). The results allowed to test dif- ferences in the regression slopes (interaction effect) and intercepts (main effects) in the two regression models (Kéry, 2010). Moreover, a regression analysis was per- formed by using the ggpubr function in R to compare the number of OTUs obtained by the two laboratories for the Lake Bourget and the data (fish number and biomass) col- lected from the traditional survey performed in Lake Bour- get in 2018 (Table S11, Figure S2). Statistical analyses were computed using R 4.03 (R Core Team, 2020). For these analyses, only the sequences identified at the species or genus level by both Lab_A and Lab_B were considered. RESULTS LAB_A analyses produced 14,600,000 sequences and, after the quality control step, an average of 250,000 se- quences per sample were obtained, excluding the negative control sample. After Obitools3 analyses, 1,500,000 se- quences were assigned, and 20 species, 9 genera, and 3 families respectively were identified. The blank sample produced less than 5000 sequences that did not find any match with vertebrate sequences present in the EMBL database, with a similarity threshold of 97%. LAB_B analyses produced an average of 320,000 se- quences per sample (excluding the template negative con- trol). After LAB_B quality control steps, an average of 270,000 sequences per sample were achieved. A total of 1,900,000 sequences after LAB_B bioinformatic analyses were successfully assigned, identifying 22 species of fish. The blank sample failed to amplify and yielded less than 500 sequences. Tissue DNA mock samples The identification success of the tissue DNA mock sample revealed good performance of both the LAB_A and LAB_B analyses, with few inconsistencies in the as- signment of OTUs to Coregonus, Esox, Cyprinus and Salmo genera (Table 1). More in detail, Coregonus lavaretus L. sequences were assigned only at the genus level by Obitools3, because the 12S rDNA fragment is not diagnostic for species within this genus, and the eco- tag script assigns a sequence to the most recent common ancestor when a sequence shows the same percentage of similarity with different species, as shown by a BLAST similarity search (Table S4): five different species were indeed identified with 100% of identity. Moreover, some of these sequences were assigned only at the family level, suggesting the presence of different genera among the reference sequences showing a level of similarity higher than 97% (as proved by a BLASTn similarity search, Table S5). The same sequence was assigned to 14 different species belonging to the Coregonus genus and one species belonging to the Stenodus genus, with a percentage of identity between 98 and 99%. The same inconsistencies are found for Cyprinus, Esox, Salmo and Silurus (Tabs S6, S7, S8 and S9). Some Squalius cephalus sequences were assigned to Leuciscus sp. because of a limited resolution power of the 12S fragment, as revealed by a Blastn search, reveal- ing a sequence similarity between 98 and 99% to both Squalius cephalus and Leuciscus leuciscus of sequences identified as Leuciscus sp. LAB_B assigned a few Core- gonus sequences to Coregonus maraena B., which is a species distributed in the Baltic Sea basin and was not included in the mock sample (Table 1). Furthermore, a wrong Salvelinus species was identified (Salvelinus fontinalis M.) instead of the expected Salvelinus alpinus L. Notably, three species present in very low proportions (Table 1) in the M14 mock sample were not detected by both the LAB_A and LAB_B procedures, namely Gobio gobio L., Esox lucius L. and Lota lota L. The comparison between the DNA proportions in- cluded in the mock samples for each species and the DNA sequences proportions retrieved using HTS for these species showed a good correlation (for both labo- ratories, Figure 2). This was also confirmed for the mock assemblage with the most complex species composition (M14). In the three mock assemblages, the individual slopes were always significant (p<0.001) and ranged be- tween 0.8 and 1.1. Moreover, both slopes and intercepts did not show significant differences (p>0.05). The mean genetic K2P distance computed among the Salmo trutta L. sequences assigned at the species level was <0.01, whereas the K2P distance calculated among the sequences identified only as Salmo genus was 0.02. No n- co mm er cia l u se on ly eDNA metabarcoding test for fish biodiversity 15 Mock communities in the fish tanks The analyses performed on the DNA collected from fish tanks (Table 2, Figure 3) showed that some species can be identified by LAB_A analyses only at the genus level be- cause still missing from the EMBL database (i.e., Barbus caninus and Salmo carpio) and, for some species, the frag- ment of the 12S marker amplified by Mifish primers is not informative (see Oncorhynchus mykiss, Table S10). More- Table 1. Species identification of mock samples made of DNA extracted from fish fin clips. For M6, M9 and M14 samples 6, 9 and 14 species were pooled respectively with different proportions of DNA (see Table S1). In bold are highlighted the assignment errors. LAB_A analysis LAB_B analysis INRAE mock samples INRAE mock samples Species present Species identified M6 M9 M14 M6 M9 M14 Abramis brama (M6, M9, M14) Abramis brama 22926 10147 13607 18950 15261 14656 Ameiurus melas (M14) Ameiurus melas 0 0 0 0 0 648 Coregonus lavaretus (M6, M9, M14) Coregonus maraena 0 0 0 14566 11089 22707 Coregonus sp. 13315 7380 21001 0 0 0 Coregoninae 2495 1263 3828 0 0 0 Cyprinus carpio (M9, M14) Cyprinus carpio 0 0 0 0 19272 4010 Cyprinidae 5452 7880 4378 0 0 0 Esox lucius (M9, M14) Esox lucius 0 18327 0 0 40949 0 Esox sp. 0 2153 0 0 0 0 Not present Leuciscus sp. 1130 449 675 0 0 0 Perca fluviatilis (M6, M9, M14) Perca fluviatilis 64823 29446 95258 69639 57379 94871 Rutilus rutilus (M14) Rutilus rutilus 84 37 230 0 0 585 Salmo trutta (M6, M9, M14) Salmo trutta 53479 31158 30666 60915 55216 40149 Salmo sp. 18439 9415 11116 0 0 0 Salmo labrax 19 0 15 0 0 0 Salvelinus alpinus (M6, M9, M14) Salvelinus fontinalis 0 0 0 22446 19829 36366 Salvelinus sp. 32463 17043 44745 0 0 0 Salmoninae 11735 6261 10338 0 0 0 Silurus glanis (M9, M14) Silurus glanis 0 0 0 0 4231 798 Silurus sp. 0 56 14 0 0 0 Squalius cephalus (M6, M9, M14) Squalius cephalus 48656 20671 28778 47551 35022 35302 Tinca tinca (M14) Tinca tinca 0 0 399 0 0 991 Table 2. Taxonomic assignment results of the metabarcoding analyses of DNA collected from fish tanks. In bold are highlighted the as- signment errors. LAB_A analysis LAB_B analysis water tank sample water tank sample Species present Species identified Barbus caninus Barbus ciscaucasicus 240 0 Carassius carassius Cyprinus carpio 0 1589 Cyprinidae 891 0 Engraulis ringens 70 0 Lepomis gibbosus Lepomis gibbosus 6394 17451 Oncorhynchus mykiss 67055 54680 Oncorhynchus sp. 13361 0 Oncorhynchus mykiss Oncorhynchus clarkii henshawi 30 0 Oncorhynchus nerka 24 0 Perca fluviatilis Perca fluviatilis 17376 47373 Salmo trutta 144596 107995 Salmo sp. 34515 0 Salmo carpio Salmo labrax 79 0 Salmoninae 3555 0 Tinca tinca Tinca tinca 6059 17533 No n- co mm er cia l u se on ly G. Riccioni et al.16 over, LAB_A analyses detected the presence of Engraulis ringens, a species often used to produce a fish meal for aquaculture. All the species present in the fish tank sample were detected by the two laboratories, except Barbus can- inus, which was not identified by LAB_B analyses even at the genus level, and Carassius carassius, whose 12S se- quence is not taxonomically informative (Table S6). The Venn diagram highlighted that 4 out of 7 species were cor- rectly identified by the two laboratories (Figure 3). The three species missed by the molecular analysis were still absent in the EMBL database or not informative. Lake Bourget samples The metabarcoding analyses performed on the environ- mental samples collected in Lake Bourget showed a good agreement between LAB_A and LAB_B (Figure 4). Only for the rare species represented with a low number of se- quences some discrepancies emerged (e.g., Barbatula bar- Figure 2. Relationship between the fraction of assigned High Throughput Sequencing (HTS) and the fractions of fish mock DNA pooled in different proportions. Each point represents 1 different species, as represented in Tab 1. The graphs refer to mock assemblages with mix of (A) 6, (B) 9 and (C) 11 species; in (C), three species with a very low amount of DNA were excluded from the analysis; see Tab 1 for details about the species included in the three graphs. In the legend, LAB_A and LAB_B refer to the two sequencing facilities. In the three mock assemblages (graphs A-C), all the regression lines were highly significant (p<0.001), whereas both slopes (around 1) and intercepts did not show significant differences. (D) Relationships between the proportions of assigned sequences detected in LAB_A and LAB_B from the analysis of fish mock DNA (r2=0.92, p<0.001). No n- co mm er cia l u se on ly eDNA metabarcoding test for fish biodiversity 17 batula L., Gasterosteus aculeatus L., Pseudorasbora sp., Oncorhynchus mykiss). The taxonomic identification al- lowed for the detection of 23 species or genera from the se- quences processed by LAB_A, and 20 species from the sequences processed by LAB_B. In total, 13 fish species were detected by both laboratories (without considering the species assigned only at the genus level; Figure 4). As for the other taxa, some species were assigned by Lab_A only at the genus level, and the analyses assigned the sequences to different species within a genus (e.g., Salmo trutta). When comparing the HTS results obtained from the three sampling stations by both LAB_A and LAB_B with the current records of species detected in the whole Lake Bourget by using the traditional sampling approach (mostly electrofishing and gill netting from 1995 to 2018), 17 and 19 species respectively out of 35 were correctly detected (Table 3, Table 4). When considering only the most recent survey, performed in 2018 with traditional gear, 13 out of 15 species were identified by the eDNA metabarcoding ap- proach (Table S11). The regression analyses performed on OTUs numbers, and traditional survey data (fish number and biomass, Table S11) showed significant R values for all the comparisons (Figure S2). The rarefaction curves constructed by using the sequences obtained from Lake Bourget samples showed that the sequencing effort could describe the diversity present in the three samples as all the three curves reached the plateau (Figure S1). By limiting the comparison only to the assigned se- quences identified at the species or genus level by both Lab_A and Lab_B, the results confirmed the high degree of comparability of HTS data (Figure 5). The Spearman correlations between the fractions of OTUs abundances Figure 3. Venn diagram of the eDNA metabarcoding taxonomic identification obtained from the samples collected from the fish tanks at FEM fishery facility. LAB_A = Fondazione Edmund Mach, LAB_B =Nature Metrics. Figure 4. Venn diagram of the eDNA metabarcoding taxonomic identification of the water samples collected in Lake Bourget. LAB_A = Edmund Mach Foundation, LAB_B = Nature Metrics. Figure 5. Relationship between the fractions of assigned se- quences obtained from the high throughput sequencing (HTS) analyses of the three environmental samples collected in Lake Bourget determined in LAB_A and LAB_B. Each point repre- sents 1 different species. The comparison is limited to the com- mon species found in the two sets of analyses. BouB = low Lake Bourget, BouM = medium Lake Bourget, BouH = high Lake Bourget. No n- co mm er cia l u se on ly G. Riccioni et al.18 computed on the total of the respective samples obtained in LAB_A and LAB_B for the three sets of samples ranged between 0.84 and 0.85 (0.001