Assessing fish diversity in small streams and ponds of the Peruvian Amazon using environmental DNA metabarcoding

TL;DR

This study uses environmental DNA to assess fish diversity in small streams and ponds in the Peruvian Amazon, showing the potential of this method for biodiversity monitoring.

Contribution

The study demonstrates the effectiveness of eDNA metabarcoding for fish community inventories in understudied Amazonian environments.

Findings

eDNA metabarcoding identified 44 MOTUs across four fish orders in small water bodies near Iquitos.

The method detected both common and elusive species, highlighting its potential for biodiversity assessment.

Incomplete DNA barcode reference libraries limit the accuracy of taxonomic assignments.

Abstract

The Amazon basin harbors exceptional fish diversity, with more than 3,500 species reported. However, this biodiversity is increasingly threatened by anthropogenic activities and climate change. The Peruvian Amazon alone is home to nearly 1,000 freshwater fish species – approximately one-third of the entire Amazon – yet significant gaps remain in our understanding of their distribution, ecology, and conservation status. Molecular approaches, particularly environmental DNA (eDNA) metabarcoding, have emerged as promising alternatives for rapid and accurate biodiversity assessment. In this study, a metabarcoding workflow targeting a fragment of the 12S gene to eDNA samples collected from small streams and ponds near Iquitos, Peru, was applied to evaluate the applicability of this approach for local fish community inventories. Water from 12 sites was filtered, and DNA was extracted, amplified, and sequenced using high-throughput Illumina technology. Bioinformatic analyses included MOTU clustering, haplotype identification, and taxonomic assignment using the Lowest Common Ancestor (LCA) algorithm. A total of 226 amplicon sequence variants (ASVs) were identified, 95 of which were assigned to fish taxa and clustered into 44 MOTUs across four orders: Characiformes, Gymnotiformes, Siluriformes, and Cichliformes. The study highlights the effectiveness of eDNA metabarcoding in detecting both common and elusive species, while also highlighting current limitations due to incomplete DNA barcode reference libraries. These findings underscore the need to expand barcode databases to fully harness the potential of eDNA for monitoring Amazonian fish biodiversity and support its use as a valuable tool for inventorying fish communities in sensitive and understudied environments.

Figures

Map of the study area in the Loreto Department, Peru.

Taxonomic assignments using the LCA algorithm with default parameters and setting the minimum percent identity to 98.0% with MEGAN6.

Hierarchical cluster of the 12 sampled sites using the Raup-Crick dissimilarity index and presence/absence of MOTUs, green lines represent rivers (largest lines) and streams (smaller lines), blue lines represent ponds.

Tables

Table 1.. Summary statistics of the similarity analysis for the 44 MOTUs assigned to a taxon.

Taxa	MOTU	Identity (%)	Nucleotide identity range	Nucleotide cover range	Total read number	Number of ASV
Specie level
Bujurquina mariae *	1	99	169	170	720	1
Crenicichla sp., Crenicichla alta* (Sexatilia alta*)	2	98–99	171–173	174	9008	1
Acestrorhynchus falcatus, Acestrorhynchus sp.	3	99	160–168	162–170	21733	3
Leporinus lacustris*, Leporinus sp.	4	98–99	166–171	169–172	6231	6
Leporinus piau*, Leporinus sp.	5	98–99	166–167	169	920	2
Schizodon fasciatus, Schizodon sp.	6	99–100	170–171	170–172	285	1
Planaltina myersi *	7	98	103	105	638	1
Astyanax bimaculatus	8	98	98	100	5957	1
Tetragonopterus argenteus	9	99	172–173	174	1094	2
Crenuchus spilurus, Crenuchus sp.	10	99–100	170–172	172	7289	1
Erythrinus erythrinus, Erythrinus sp.	11	98–100	160–167	162–167	30846	7
Hoplerythrinus unitaeniatus, Hoplerythrinus sp.	12	99–100	163–166	165–166	43894	4
Hoplias malabaricus	13	98	163–164	166–167	10478	3
Carnegiella strigata	14	100	163–168	163–168	12382	2
Prochilodus harttii*, Prochilodus lineatus*, Prochilodus costatus*, Prochilodus nigricans, Prochilodus magdalenae*, Prochilodus reticulatus*, Prochilodus argenteus*	15	98–100	163–171	165–171	7049	1
Triportheus angulatus	16	100	170	170	642	1
Electrophorus electricus *	17	100	167	167	1690	1
Electrophorus voltai*, Electrophorus sp.	18	98–100	164–167	167	1138	1
Gymnotus carapo	19	98–100	167–172	170–172	42388	9
Apteronotus sp., Apteronotus albifrons	20	99	171	172	8397	1
Brachyhypopomus verdii, Brachyhypopomus sp.	21	98–99	164–167	165–168	9417	4
Microsternarchus bilineatus *	22	98	170	173	504	1
Megalechis picta	23	98–99	174–175	177	654	1
Ancistrus trinitatis *	24	99	169	171	2799	1
Hypostomus sp., Hypostomus ancistroides*	25	98	168	171	683	1
Farlowella oxyrryncha, Farlowella knerii, Farlowella schreitmuelleri*, Farlowella paraguayensis*, Farlowella hahni*, Farlowella reticulata*, Farlowella smithi	26	98–100	170–173	173–174	1295	1
Rhamdia quelen	27	98–99	105–171	106–174	42940	2
Genus level
Aequidens: Aequidens sp.	29	98	162–167	165–170	24193	3
Hypselecara: Hypselecara coryphaenoides* or Hypselecara temporalis	30	99	168	170	19004	3
Crenicichla: Crenicichla sp., Crenicichla alta* (Sexatilia alta*), Crenicichla frenata* (Sexatilia frenata*)	31	98–99	168–171	171–174	2399	2
Moenkhausia sp.	32	99	166	168	7778	1
Hoplias: Hoplias malabaricus, Hoplias aimara*, Hoplias sp.	33	98–99	160–165	163–167	17168	2
Semaprochilodus sp.	34	98	169	172	398	1
Brachyhypopomus: Brachyhypopomus sp.	35	99–100	166–168	169	3026	3
Semaprochilodus: Semaprochilodus sp., Semaprochilodus taeniurus	36	99–100	169–172	170–172	14435	1
Eigenmannia sp.	37	98	173	176	6965	1
Eigenmannia: Eigenmannia sp.	38	99	167–168	169	5181	2
Sternopygus sp.	39	98–99	165–172	168–174	42954	4
Peckoltia sp.	40	98	168	171	5578	2
Family level
Serrasalmidae: Myloplus rubripinnis*, Myloplus sp., Myleus tiete	41	99–100	172–173	173–175	59667	1
Suborder
Characoidei: Metynnis maculatus, Brycon melanopterus	42	99–100	166–168	168	6253	3
Siluroidei: Pimelodella sp., Pimelodus mysteriosus*, Pimelodus blochii, Pimelodus cf. blochii, Pimelodus pohli*, Pimelodus yuma*, Pimelodus argenteus*, Pimelodus coprophagus*, Pimelodus sp.	43	98–99	168–172	171–175	710	1
Subdivision
Percomorphaceae: Australoheros facetus*, Aequidens metae*, Pomadasys ramosus*, Centropomus undecimalis*, Bairdiella ronchus*, Centropomus parallelus*, Pachyurus adspersus*, Micropogonias furnieri*, Cichlasoma sp.	44	98–100	160–169	163–170	97410	3
Supercohort
Clupeocephala: Serrasalmus sp., Pygocentrus nattereri, Pterophyllum scalare, Pygocentrus piraya*, Serrasalmus rhombeus, Serrasalmus brandtii*, Serrasalmus serrulatus, Serrasalmus eigenmanni, Serrasalmus elongatus, Pristobrycon striolatus	45	98–100	160–170	162–171	8201	1
Not assigned						131

Full text

Introduction

With more than 3,500 fish species reported to date, the Amazon Basin is the most species-rich river system in the world (Jézéquel et al. 2020). However, its biodiversity is among the most threatened (Malhi et al. 2008; Beuchle et al. 2021). Since the 1970s, anthropogenic activities, including deforestation, infrastructure development, mining, oil extraction, pollution, and overharvesting, have increased exponentially to meet the demands of a growing population and international markets (Beuchle et al. 2021; Albert et al. 2025). Moreover, the latest IPCC report highlights that temperatures will continue to rise in the Amazon (Lee et al. 2023), leading to major changes in water circulation across the continent (Espinoza et al. 2021). Climate change is currently driving major species reshuffling across the globe (Devictor et al. 2012; Pecl et al. 2017). However, these dynamics remain poorly documented in the Amazon due to significant gaps in biodiversity knowledge.

Aquatic ecosystems are keys components of the Amazon Basin (Castello et al. 2013; Albert et al. 2025), and fish are essential for local populations, as they represent a substantial source of animal protein (Adams et al. 2009). However, due to their high diversity and dramatic phenotypic changes during development, identifying fish species is a challenging task. In this context, taxonomic knowledge gaps related to species identity (i.e., the Linnean shortfall) and species distribution (i.e., the Wallacean shortfall) hinder accurate monitoring of Amazonian fish species. However, the development of molecular approaches, such as DNA barcoding for species delimitation and specimen identification, has greatly improved the accuracy and speed of ichthyological inventories in the Amazon, opening new perspectives for monitoring efforts (Pereira et al. 2013; Guimarães et al. 2018; Machado et al. 2018; de Freitas et al. 2024).

More than 1,000 fish species have been reported from the Peruvian freshwater ecosystems (Ortega et al. 2012) representing nearly one-third of all fish species found in the Amazon (Jézéquel et al. 2020). However, the range distribution, ecology, and conservation status of most of these species remain largely unknown. These major knowledge gaps are likely due to the complexity and vastness of the Amazon’s aquatic environments, which make sampling individuals a challenging and time-consuming task. In this context, inventory methods adapted to these environments are required to accelerate research. Environmental DNA (eDNA) has opened new perspectives in this field, and metabarcoding – the simultaneous identification of numerous individuals through high-throughput sequencing – has the potential to become the fastest and most effective method for inventorying species across extensive geographical scales (Taberlet et al. 2018). One of its main strengths is its potential effectiveness to detect elusive species that are difficult to observe due to environmental conditions limiting visibility (such as currents, turbidity, and depth) or because of their ecological preferences (e.g., cryptobenthic species) (de Santana et al. 2021; Ruppert et al. 2019). To date, eDNA metabarcoding has increasingly been applied to assess fish diversity in the Amazon Basin, particularly using the 12S rRNA gene region (Cantera et al. 2019, 2022a, b, 2023; Cilleros et al. 2019; Bevilaqua et al. 2020; de Santana et al. 2021; Batista et al. 2022; Coutant et al. 2023; Condachou et al. 2024; Jackman et al. 2024; Martinelli Marín et al. 2024; Timana‐Mendoza et al. 2025), the Cytochrome oxidase I gene (Mariac et al. 2022; Timana‐Mendoza et al. 2025) and the 16S region (Bevilaqua et al. 2020), have also been explored. Most of these studies have shown that eDNA metabarcoding is a promising approach for inventorying the Amazonian ichthyofauna, despite reference libraries still being incomplete.

Here, we applied an eDNA metabarcoding workflow to samples collected in small streams and ponds near Iquitos, in the Peruvian Amazon. Using a fragment of the 12S gene region, we aimed to inventory fish communities and evaluate applicability of this method in this part of the basin. We further discuss the benefits, limitations, and implications of eDNA metabarcoding for improving biodiversity assessments in Amazonian freshwater ecosystem.

Materials and methods

Study area

The study was conducted in the Loreto Region of the Peruvian Amazon, around two towns: Jenaro Herrera on the right bank of the Ucayali River, and Nauta, on the left bank of the Marañón River (Fig. 1). Water samples were collected from ten sites in Jenaro Herrera area and two sites in the Nauta area. At each site, the highest volume possible was filtered, with an average volume of 22.5 L. However, standard deviation was high (SD ± 9.2 L) due to the occurrence of clogging. The sampling sites consisted of networks of smalls forest streams (~ 0.5–6 m wide) and ponds (~ 20–3,000 m^2^) in terra firme forest. Water type across all sites ranged from clear to slightly blackish water, with no suspended sediment except at site 12, which showed a moderate sediment load.

Map of the study area in the Loreto Department, Peru.

eDNA sampling and extraction

Water samples from 12 sites were filtered using Waterra cartridge with a 0.45 µm pore size: ten sites in the region of Jenaro Herrera and two in the region of Nauta (Fig. 1). The filtration system consisted of tubing of 6 and 8 mm, and a water pump with a micro diaphragm (SEAFLO 12V 3,8 L/min, 40 psi) powered by a 12 V ion battery of a capacity of 7.5 mAh. To monitor potential contaminations related to pre-PCR steps, an additional cartridge was used in the field to filter 21 liters of bottled water. After filtering, the remaining water in each cartridge was expelled by introducing air into the system, and the cartridge was then filled (approx. 80 ml) with 1× Longmire buffer (Tris–HCl 0.1 M, EDTA 0.1 M, NaCl 0.01 M and N-lauroyl sarcosine 1% with pH 7.5–8). Each cartridge was labeled with a unique code and stored at 4 °C until DNA extraction.

The eDNA contained within the cartridges was extracted using the following protocol: the cartridges were placed in a 56 °C incubator for 2 h. During the incubation, the cartridges were shaken at least twice. The contents of each cartridge were then poured and weighed into a biological sample collection bottle and divided equally into three 50-mL Oak Ridge High-Speed PPCO Centrifuge tubes (Nalgene^TM^). The tubes were centrifuged at 15,000 × g for 15 min at 6 °C. After centrifugation, the supernatant from each tube was carefully pipetted off, leaving ~ 12 g of material in each tube. Twenty-seven milliliters of absolute ethanol and 1.2 mL of 3 M sodium acetate were added to the tubes, which were placed at -20 °C overnight. The tubes were subsequently centrifuged at 15,000 × g for 15 min at 6 °C, and the supernatants were carefully discarded by tilting. Next, 2 × 720 µL of ATL Buffer (Tissue lysis buffer, Qiagen) was added to each tube, and each was gently vortexed to dislodge the “clean” pellet from the bottom. The contents of each tube were carefully pipetted, avoiding the “dirty” fraction, and transferred into 2 mL tubes. To each tube, 20 µL of proteinase K (20 mg/µL) was added, and the samples were incubated for 2 hours at 56 °C in an Eppendorf ThermoMixer with shaking at 600 rpm. For each tube, 800 µL of the mixture was pipetted twice and distributed into two tubes of 2 mL, to which 250 µL of SB buffer was added. The tubes were vortexed and briefly centrifuged for ~ 5 s. The supernatants were then loaded onto columns corresponding to step 7 of the NuceloSpin Soil kit (Macherey-Nagel), and the extraction protocol was continued accordingly. In the final stage of the extraction, 50 µL of SE buffer pre-warmed to 56 °C was added on the membranes to elute the DNA. At the end of the extractions, 76 samples were obtained – six for each cartridge (6 × 13). To prevent cross-contamination, all extractions were conducted on a laboratory bench dedicated to eDNA analyses, sterilized with DNA away prior to each extraction. Pipettes were UV-sterilized before use, filtered tips were systematically used, and gloves were changed between each cartridge.

The quality and concentration of the DNA were measured using a nanodrop (thermos scientific), and the double-stranded DNA (dsDNA) were quantified with a Qubit 4 (Invitrogen).

DNA amplification, libraries construction and NGS sequencing

For each cartridge, three of the six aliquots with the highest dsDNA concentrations were selected for PCR amplification. Reactions were performed using AmpliTaq Gold 360 Master Mix (Applied Biosystem) and 12S Mifish primers (MiFish-F 5’_GTCGGTAAAACTCGTGCCAGC_3’, MiFish-R 5’_CATAGTGGGGTATCTAATCCCAGTTTG_3’; Miya et al. 2015). An extension sequence was added to 5’ end of both the forward and the reverse primers: TCGTCGGCAGCGTCAGATGTGT ATAAGAGACAG and GTCTCGTGGGCTCGGAGATGTGTATAAGAGA CAG, respectively. Extensions were used for library indexing. Between the extension sequences and the primer sequence, one to three nucleotides were added to prevent signal saturation during Illumina Miseq sequencing. Each PCR reaction had a total volume of 20 μL and contained 10 μL of AmpliTaq Gold 360 Master Mix, 0.16 μL of bovine serum albumin (BSA), 5.84 μL of water, 2 μL combined forward and reverse primer mix (5 μM) and 2 μL of template eDNA. The program consisted of an initial Taq activation at 95 °C for 10 min, followed by 45 cycles of 95 °C for 30 s (denaturation), 60 °C for 30 s (annealing), and 72 °C for 60 s for extension, with a final extension step at 72 °C for 7 min.

Libraries were prepared using a two-step PCR protocol (Galan et al. 2018). In the first PCR step, a total of 168 samples were amplified, corresponding to each cartridge (12 sites + 1 negative control), three extraction aliquots, and four PCR replicates (13 cartridges × 12 replicates + 12 negative controls). In the second PCR step, each library was indexed using unique dual-index combinations to exclude chimeric amplicons more effectively during bioinformatics processing. Library was sequenced with 150 bp paired-end reads on an Illumina Miseq platform. The second PCR step and Illumina Miseq sequencing was performed by the Genseq platform (CNRS, Montpellier, France). Illumina Miseq sequencing run are available as SRA in NCBI GenBank (BioProject ID: PRJNA1391834).

Metabarcoding workflow

Bioinformatic analyses were performed on demultiplexed data using the JAMP pipeline developed by Elbrecht et al. (2018). This workflow uses the library “JAMP” in R, along with the tools VSEARCH (Rognes et al. 2016), USEARCH (Alloui et al. 2015) and CUTADAPT (Martin 2011). The main steps of the analysis included, reads merging, primers trimming, maximum expected error filtering, normalization to the same sequencing depth (to allow comparisons between sites), fragment length filtering, reads filtering to identify ASV (i.e., haplotype) including reads denoising, Molecular Operational Taxonomic Unit (MOTU) delimitation through ASV clustering based on a threshold approach (i.e., 97% similarity), and abundance-based filtering within and between libraries.

The sequence depth was normalized to 20000 reads per library, and fragment lengths were filtered to retain reads between 150 and 200 base pairs. No threshold filtering abundance was applied at the individual library level in order to be less conservative; instead, filtering was performed across libraries for each site (12 libraries per site). Haplotypes were retained if they were detected in at least six of the 12 libraries (≥50%) and discarded otherwise. Parameters used in the Denoise function were as follows: minsize = 20, minrelsize = 0, OTUmin = 0, minhaplosize = 0, withinOTU = 0, eachsampleOTUmin = 0, minHaploPresence = 6, minOTUPresence = 1, and other parameters were set to default values.

All haplotypes identified within each MOTU were compared against the NCBI-nr database using BLASTn (Chen et al. 2015), and the results were formatted using DIAMOND (Benson et al. 2013; Buchfink et al. 2015). BLASTn outputs were then used for taxonomic assignment with the Lowest Common Ancestor (LCA) algorithm implemented in Megan6 (Huson et al. 2016). In Megan, taxonomic identification is not based on the single best hit; instead, each read is assigned to the lowest taxonomic node that includes all valid hits above the defined thresholds (e.g., bit score, percent identity). Megan6 was run with default parameters and a minimum percent identity threshold of 98.0%; haplotype with blast identity scores below 98% were considered unassigned. ASVs collected here are made available in a fasta file in Suppl. material 3.

Spatial structure of fish communities

We examined the spatial structure of fish assemblages in the study area by performing a hierarchical cluster analysis based on the Raup-Crick dissimilarity index (Raup and Crick 1979). This index was preferred because it is a simple and explicit way to measure dissimilarity in species composition between sites (0 = all species are shared, 0.5 = half of the species are shared, 1 = no species are shared) while adjusting for heterogeneous species richness between sites. The dissimilarity matrix among sites was calculated using the presence/absence of MOTUs using the raupcrick function from the R package vegan (Oksanen et al. 2022). The hierarchical clustering was then performed using the R package stats 3.1.2 (R Core Team 2025).

Results

Illumina Miseq sequencing generated a total of 5,864,056 reads across 168 libraries. From the 12 sampled sites, 226 ASVs were detected (Suppl. material 3), of which 95 could be assigned to the species level with the 98% threshold. Of these, all but one corresponded to fish species and were clustered into 44 MOTUs. Of these, 27 were identified at the species level, 12 at the genus level, one at the family level, two at the suborder level, and two at higher taxonomic ranks (Table 1, Fig. 2).

Taxonomic assignments using the LCA algorithm with default parameters and setting the minimum percent identity to 98.0% with MEGAN6.

The 44 MOTUs were assigned to four orders: Characiformes (45%), Gymnotiformes (23%), Siluriformes (16%), and Cichliformes (14%), with an additional 2% corresponding to an unresolved basal node (Fig. 2). The number of MOTUs detected per site ranged from one (site 7) to 19 (site 12). The most frequently detected species across sites was Gymnotus carapo, found in five of 12 sites. In terms of reads counts, Hoplerythrinus unitaeniatus (43,894 reads), Rhamdia quelen (42,940 reads), and Gymnotus carapo (42,388 reads) were the most abundant taxa (Table 1).

The species with the highest number of ASVs were Gymnotus carapo (9 ASVs), Erythrinus erythrinus (7 ASVs), and Leporinus piau (6 ASVs) (Suppl. material 1). For Gymnotus carapo, four ASVs were found exclusively at site 6, one at site 1, and one ASV was shared between sites 9 and 11. For Erythrinus erythrinus, six ASVs were found at site 11, three at site 1, one ASV was shared among sites 1, 2, 10, and 11, and another ASV was shared between sites 1 and 11. All six ASVs identified for Leporinus lacustris were found at site 12.

Site 12, in addition to being the richest in terms of species among all the studied sites, exhibited the most dissimilar fish community (Fig. 3). The other sites are clustered into three main groups: one cluster comprising sites 8 and 10, another one with sites 3, 4, 5, and 7, and a final cluster with sites 1, 2, 6, 9, and 11 (Fig. 3). Sites 8, 10, and 12 correspond to the largest tributaries of the 12 sites, measuring approximately 3 to 6 meters wide, with site 10 being a tributary of site 8. Sites 3, 4, 5, and 7 correspond to ponds; sites 4 and 5 were sampled from the same pond. Except for site 9, which is the smallest pond sampled, sites 1, 2, 6 and 11 correspond to the smallest streams, each less than 1 m wide.

Hierarchical cluster of the 12 sampled sites using the Raup-Crick dissimilarity index and presence/absence of MOTUs, green lines represent rivers (largest lines) and streams (smaller lines), blue lines represent ponds.

Discussion

To our knowledge, this is the first study to apply environmental DNA metabarcoding to inventory fish communities specifically in small terra firme streams (<1 m wide) and ponds (20–3,000 m^2^) of the Peruvian Amazon. Across all sites studied, we observed a dominance, in term of number of species, belonging to the orders Characiformes, Gymnotiformes, and Siluriformes. These observations differ slightly from the overall Amazonian ichthyofauna, which is mainly represented by the orders Characiformes, Siluriformes and Cichliformes (Dagosta and De Pinna 2019). Other authors have reported different results based on samplings methods using both individual collections and eDNA (Zuanon et al. 2015; Jackman et al. 2021; Batista et al. 2022). These differences can be explained by the uneven distribution of the orders that make up the Amazonian ichthyofauna throughout the watershed. Variation in distribution is primarily influenced by the geological and hydrological history of the Amazon basin (Cassemiro et al. 2023; Hubert and Renno 2006), as well as the habitat heterogeneity and the ecological adaptations or preferences of certain groups to specific environments (Oberdorff et al. 2019). However, it is interesting to note the omnipresence and dominance of Characiformes in Amazonian environments, even in small streams. The sites sampled were forest streams on terra firme, characterized by low conductivity and a pH below 7. It has been shown that several genera of Gymnotiformes are associated with such environments, including Gymnotus, Brachyhypopomus, Hypopygus, Microsternarchus, Gymnorhamphichthys, Sternopygus and Eigenmanna (Albert and Crampton 2005). In our study, we detected all these genera except Hypopygus and Gymnorhamphichthys, as well as Electrophorus and Apteronotus.

In general, the genera and species identified in this study are known to occur in Peru and are expected in this type of environment. However, several species detected in our dataset – such as Leporinus lacustris, Leporinus piau, Planaltina myersi, Electrophorus electricus, Electrophorus voltai and Ancistrus trinitatis – have not previously been recorded in Peru. Some of these records may reflect recent taxonomic revisions (e.g., Electrophorus), but they are more attributable to misidentifications in the NCBI database and potential errors in the phylogenetic assignments performed by MEGAN during Lowest Common Ancestor (LCA) classification.

Additional factors contributing to uncertain assignments include the absence of 12S barcodes for many local species and the low genetic divergence among congeneric taxa. This is well exemplified by the spurious detection of Ancistrus trinitatis, a species originally described from the island of Trinidad and belonging to the taxonomically complex genus Ancistrus, which includes numerous closely related and poorly defined species. Similarly, Leporinus piau and L. lacustris were detected in our dataset, although their known distributions are restricted to northeastern Brazilian coastal rivers and the Paraná River Basin, respectively. These improbable records most likely reflect incomplete or misannotated reference libraries, a common limitation in Amazonian eDNA studies. Taken together, these observations emphasize the need for expanding and curating regional molecular reference databases to improve species-level resolution and underscore the importance of interpreting eDNA-based taxonomic assignments with caution (Blackman et al. 2023; Gold et al. 2021; Keck et al. 2023). For instance, of the 226 ASVs identified, 131 could not be assigned to any taxon with an identity ≥ 98% (Suppl. material 2), but were assigned to a sequence with ≥ 98%. This was due to the absence of 12S reference barcodes for these particular taxa.

Besides, the Linnean shortfall – i.e., the gap between described species and actual biodiversity – further limit the effectiveness of eDNA metabarcoding, which remains constrained to a relatively small set of well-known species. According to Marques et al. (2021) and the GAPeDNA v. 1.1.2 platform (https://shiny.cefe.cnrs.fr/GAPeDNA/), only 11% (254 of 2273) of Amazonian fish species have a 12S reference barcode (MiFish), and 40–60% of species classified as vulnerable by the IUCN lack any reference sequence. This lack of 12S reference information is likely one of the main reasons for the relatively low richness of Siluriformes recorded in our study. Many siluriform species occurring in the Amazon probably lack 12S barcode data, which limits their detection and assignment in metabarcoding analyses.

During our study, conducted as part of a project assessing the reliability of eDNA for characterizing intraspecific genetic diversity of Apistogramma agassizii, individuals of A. agassizii were collected from 11 of the 12 sites. However, assignments with ≥98% identity did not confirm the presence of A. agassizii in the study area. Among the 131 unassigned ASVs, some were attributed to A. agassizii, but the identity percentages did not exceed 94%. Sanger sequencing of the 12S fragment of A. agassizii individuals collected in the study area (A. agassizii Sp1, cf.: Estivals et al. 2023) confirms the presence of the species, matching with 100% identity to one of the previously unassigned ASVs. This example perfectly illustrates the impact of taxonomic gaps in the reference libraries and highlights the need to generate reference barcodes to fill those gaps and ensure the full effectiveness of eDNA metabarcoding approaches.

Finally, our study supports previous findings suggesting that eDNA effectively captures the structure of fish assemblages in the Amazon (Cantera et al. 2022a; Mariac et al. 2022), offering some new perspectives for the characterization of the Amazonian biotas. The sites corresponding to the largest tributaries (8, 10, and 12) host the highest number of species – an observation previously reported by Stegmann et al. (2019). However, the opposite trend is observed at the scale of the Amazon basin due to the complex geological and hydrographical history of the basin (Carvajal-Quintero et al. 2019; Oberdorff et al. 2019). At the more local scale of our study, larger rivers likely support higher species richness because they provide more space, a greater variety of habitats, and increased food availability. Interestingly, fish communities in our study also grouped according to environmental type – lotic versus lentic systems – consistent with patterns observed at broader scales (Mariac et al. 2022).

Conclusions

Our study demonstrates that the fragment of the 12S gene amplified with the MiFish primers is highly specific and effective for inventorying the Amazonian ichthyofauna. The detection of numerous ASVs (haplotypes) within certain species suggests that the 12S MiFish fragment may also be valuable for intraspecific genetic studies. Overall, eDNA metabarcoding presents an interesting alternative to traditional individual sampling, especially in sensitive environments such as small rivers in the Amazon. However, as highlighted by most studies, a significant gap remains in the barcode reference libraries. The main challenge in the coming years will not be the technical ability to conduct studies eDNA metabarcoding studies in the Amazon, but rather to bridge the barcode gaps across Amazonian countries, which currently limit the full effectiveness of eDNA approaches.