Haplotype-level analysis of environmental DNA metabarcoding revealed the biogeography and phylogeography of freshwater fishes in Korean Peninsula

Abstract

Environmental DNA (eDNA) metabarcoding is a cost-effective, sensitive, and minimally invasive tool for monitoring fish diversity in freshwater ecosystems. To evaluate its suitability for assessing fish biogeography and phylogeography in Korean streams, we analyzed fish assemblages in three streams—Gilan, Seom, and Oshipcheon—located in different Korean biogeographical subdistricts. Over two sampling campaigns at nine sites, a total of 107 amplicon sequence variants (ASVs) were identified, representing 20 families, 46 genera, and 76 species. Among the 76 detected species, 29 were endemic to freshwater habitats in South Korea and played a significant role in shaping fish communities across the surveyed streams. We identified potential sources of six translocated species to the eastern region by examining their ASVs. After excluding these potential translocations, no endemic fish species were simultaneously detected in all three streams, and the phylogeography of endemic fish species Odontobutis platycephala, Coreoperca herzi, Microphysogobio yaluensis, Coreoleuciscus splendidus, Nipponocypris koreanus, and Koreocobitis rotundicaudata was clearly observed across the three biogeographical regions. Additionally, analysis of relative abundance and presence-absence data of fish communities yielded comparable β-diversity metrics reflecting spatiotemporal variation. The fish communities in the streams exhibited distinct groups, with Gilan showing a closer relationship to Seom than to Oshipcheon, consistent with well-known allopatric studies of freshwater fishes in the Korean Peninsula. Interestingly, we did not find any differences in fish assemblages among the three mesohabitat types (riffle, run, and pool), regardless of the sampling site. Our results highlight the potential of water-derived eDNA metabarcoding for detecting endemic fish, unraveling the biogeography and phylogeography of allopatric populations of freshwater fishes, and providing genetic forensic tools to estimate the original sources of translocated species.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-36043-x.

Introduction

The biogeography and phylogeography of freshwater fishes are influenced by intricate ecological, geological, and evolutionary processes. These dynamics are notably present in regions with distinctive zoogeographical patterns and an array of biodiversity, such as the Korean Peninsula. The native freshwater fishes of Korea, which comprise approximately 30% of the known species, provide essential insights into the impact of historical and contemporary events on biodiversity¹. Despite their importance, these native species are increasingly at risk due to habitat degradation, climate change, and anthropogenic activities, such as species translocation and habitat alteration². These disturbances present problems for biodiversity conservation and require accurate tools to evaluate the biogeographical distributions as well as the population structure.

Typically, the conventional methods for investigating the diversity of freshwater fishes, such as electrofishing and fyke net surveys, are labor-intensive, costly, and disruptive to habitats, which restricts their practicality for large-scale biodiversity assessments³. Additionally, tissue collections were required for assessing phylogeography for DNA source and it was problem if targeted the endemic fishes which usually appeared in low abundance. In this context, environmental DNA (eDNA) metabarcoding has emerged as a transformative tool for ecological and evolutionary studies⁴. eDNA metabarcoding facilitates the identification of species and their genetic markers from trace DNA recovered from environmental samples, such water or soil⁵. In addition to species detection, recent advancements in bioinformatics have enhanced the capability of eDNA metabarcoding to identify haplotypes⁶, creating opportunities for phylogeographical and population genetic research with remarkable accuracy to examine evolutionary mechanisms and population connections^7,8. However, applications of metaphylogeographic analyses to freshwater fishes remain limited, highlighting the need for studies exploring the phylogeographic structure of riverine taxa through eDNA metabarcoding.

The utilization of haplotype-level eDNA metabarcoding holds significant potential for areas with intricate biogeographical backgrounds, illustrated by Korea. The Korean Peninsula are categorized into three principal zoogeographical subdistricts, including Northeast Korea Subdistrict (NKS), South Korea Subdistrict (SKS), and West Korea Subdistrict (WKS)⁹. Each subdistrict contains distinct fish assemblages and endemic species influenced by previous climatic events and geological barriers age¹⁰. Species such as Squalidus multimaculatus (Hosoya & Jeon, 1984), Iksookimia koreensis (Kim, 1975), and Niwaella multifasciata (Wakiya & Mori, 1929) are confined to particular subdistricts, illustrating patterns of allopatric divergence and local adaption². Nevertheless, the genetic architectures of these populations are progressively compromised by human activities, such as artificial stock augmentation initiatives¹¹. Even though fish eDNA metabarcoding has been applied in Korean rivers, the previous studies have primarily focused on species diversity and ecological dynamics of fish communities^12,13.

This study aims to assess the potential of haplotype-based fish eDNA metabarcoding for addressing various key questions within Korea’s riverine systems, including fish assemblages, biogeographical distributions, and phylogeography of native species. To achieve this, three small streams was selected, namely Seom, Gilan, and Oshipcheon, which represent three distinct geographical subdistricts in Korea. The Seom stream is a significant tributary of the lower Han River in the WKS, which encompasses the largest river basin in South Korea and flows towards the Yellow Sea¹⁴.The Gilan stream is located within the northern riverine system of the Nakdong River in the SKS and is characterized by a high frequency of Korean endemic species¹⁵. In contrast, the Oshipcheon stream is a typical stream in the NKS, flowing into the East Sea and characterized by small, individual channels with high slopes¹⁶. This study will contribute to the scientific conservation and management of native fishes inhabiting diverse riverine systems in Korea.

Results

General results of fish metabarcoding

Following the denoising procedure, a total of 4,668,380 non-chimeric reads (58.26% of total reads, Supplementary Table S1) and 918 amplicon sequence variants (ASVs) were obtained. After implementing the LULU algorithm for abundance curation and removing non-fish reads (n = 56,831), 4,611,549 reads (98.78% of the total) and 107 ASVs were identified as reliable fish reads and haplotypes, respectively. These ASVs represented ten orders, 20 families, 46 genera, and 76 species. Among the 107 fish ASVs, 103 exhibited 97% or higher sequence identity to the reference database, with 69 ASVs showing 100% identity (Supplementary Table S2), indicating a substantial coverage of reference sequences in the database. Only four ASVs (11,545 reads) with sequence identities ranging from 95% to 97% were assigned to the genus level (Sarcocheilichthys, Pseudorasbora, Silurus, and Zacco). The order Cypriniformes had the highest number of families (n = 7) and species (n = 49), followed by Siluriformes with three families and nine species (Fig. 1). The family Leuciscidae accounted for the most abundant read numbers, representing 24.63% of the total reads, while the family Gobionidae had the highest number of species (n = 20). Regardless of the stream, Rhynchocypris oxycephala was the most abundant species (RRA 16.88%), followed by Pungtungia herzi (Herzenstein, 1892) (14.44%) and Zacco platypus (Temminck & Schlegel, 1846) (11.88%) as the representative species inhabiting small streams (Fig. 1B).

Fig. 1.

Read proportion plot of freshwater fish taxa (A) family proportions in March and August; (B) relative contribution of several taxa levels (order, family, and species) showed by Sankey diagram. The thickness of the link represents the read counts of corresponding taxa.

Across the three investigated streams, a consistent presence of seven fish species was observed, encompassing Coreoleuciscus splendidus (Mori, 1935), Silurus asotus (Linnaeus, 1758), Misgurnus anguillicaudatus (Cantor, 1842), Nipponocypris koreanus (Kim, Oh & Hosoya, 2005), Pungtungia herzi, Rhynchocypris oxycephala (Sauvage & Dabry de Thiersant, 1874), and Zacco platypus (Fig. 2). Remarkably, the latter quartet of species maintained a ubiquitous presence across all sample collection sites. Seom demonstrated a superior ichthyofaunal diversity with a species count of 49, whereas Gilan and Oshipcheon presented notably lesser numbers, both with a count of 28 species. The greatest overlapping species count was observed between Gilan and Seom (9 species), while the smallest was between Oshipcheon and Gilan (2 species). Seom displayed the highest number of unique species (28 species), succeeded by Oshipcheon (14 species) and Gilan (10 species) (Fig. 2A). Of the 75 identified species, 61 were consistent in both periods of sampling (Fig. 2B). However, eleven species (Rhinogobius giurinus (Rutter, 1897), Rhynchocypris lagowskii (Dybowski, 1869), Plecoglossus altivelis (Temminck & Schlegel, 1846), Acheilognathus majusculus (Kim & Yang, 1998), Acheilognathus rhombeus (Temminck & Schlegel, 1846), Pseudobagrus koreanus (Uchida, 1990), Tachysurus fulvidraco (Richardson, 1846), Micropterus salmoides (Lacepède, 1802), Sarcocheilichthys sp., Oryzias sinensis (Chen, Uwa & Chu, 1989), and Misgurnus anguillicaudatus) were only detected in August, and three species (Oncorhynchus keta (Walbaum, 1792), Pseudorasbora sp., and Pseudorasbora parva (Temminck & Schlegel, 1846)) were exclusive to March. A total of twenty-nine endemic species were recognized, their abundances spanning 0.01% to 5.77%. The highest proportions of these endemic species were found in Gilan (46.43%, 13 species), then Seom (41.67%, 20 species), and finally Oshipcheon (35.71%, ten species) (Fig. 2A). Several endangered species as classified by the National Institute of Biological Resources of Korea (NIBR), including Koreocobitis naktongensis (Kim, Park & Nalbant, 2000), Phoxinus phoxinus (Linnaeus, 1758), Rhodeus pseudosericeus (Arai, Jeon & Ueda, 2001), Gobiobotia brevibarba (Mori, 1935), Gobiobotia macrocephala (Mori, 1935), and Cottus hangiongensis (Mori, 1930), were identified. Additionally, the presence of two invasive species, Micropterus salmoides (in both Seom and Gilan) and Oncorhynchus mykiss (Walbaum, 1792) (in Oshipcheon) was recorded. Two species, Tribolodon hakonensis (Günther, 1877) and Culter Alburnus (Basilewsky, 1855), were identified in Seom Stream, where those species have never been recorded.

Fig. 2.

Spatiotemporal presence-absence profiles of 75 freshwater fishes correspond to (A) stream by Upset plot, (B) sampling time by Venn diagram, and (C) stream sections by Heatmap plot. Samples of mesohabitat were merged in respective sections of each stream for heatmap analysis. OP, SM, and GN denotes Oshipcheon, Seom, and Gilan, respectively.

The heatmap analysis elucidated that the ichthyofaunal compositions in each stream were grouped distinctly from the others. A comparative analysis of the streams revealed a greater resemblance in the fish assemblages between Seom and Gilan than those of Oshipcheon. The presence of species such as Cyprinus carpio (Linnaeus, 1758) and Hemibarbus longirostris (Regan, 1908), which were designated to Seom, suggested that some of our sampling locations within Seom do not epitomize the far upper reaches of the Han riverine system. Notably, the Oshipcheon was characterized by the presence of the diadromous species: Tribolodon hakonensis and three salmonid species, Oncorhynchus mykiss, Oncorhynchus masou (Brevoort, 1856), and Oncorhynchus keta (Fig. 2C). In contrast, the fish assemblage in Gilan was distinguished by the presence of Koreocobitis naktongensis, Niwaella multifasciata, and Tanakia latimarginata (Kim, Jeon & Suk, 2014).

Fish diversity and community profiles

The alpha diversity indices demonstrated lower average values in March (20.37 ± 9.09 for ASVs and 16.85 ± 8.30 for species) as opposed to the values recorded in August (26.14 ± 12.51 for ASVs and 21.78 ± 10.82 for species) (Fig. 3A). A pronounced species richness was observed in the stream of Seom, showing 34.83 ± 6.22 for ASVs and 30.61 ± 4.46 for species. Conversely, both Gilan and Oshipcheon streams presented comparably low species richness, recording 17.56 ± 7.80 and 17.39 ± 8.96 for ASVs, and 14.17 ± 5.73 and 13.17 ± 6.90 for species, respectively (Fig. 3B). The lowest ASV and species richness were observed in the upper reach of Oshipcheon during March (5.33 ± 0.58 and 3.33 ± 0.58, respectively), while the highest numbers of ASV (43 ± 1.73) and species (35.67 ± 1.15) were detected in the middle section of Seom during August. Among the three examined streams, Seom showed the highest alpha diversity estimates (2.07 ± 0.39 in Shannon and 0.77 ± 0.11 in Simpson) compared to Gilan (1.33 ± 0.53 in Shannon and 0.60 ± 0.21 in Simpson) and Oshipcheon (1.25 ± 0.74 in Shannon and 0.55 ± 0.26 in Simpson) (Fig. 3B).

Fig. 3.

The diversity estimates (number of ASV and species, Shannon, and Simpson index) of freshwater fish communities inferred from eDNA metabarcoding (A) by month and (B) by stream. Means and medians are presented as solid black dots and horizontal solid lines within each box, respectively. Asterisks indicate the significance level (p) for each comparison by pairwise Wilcoxon rank-sum tests: * p < 0.05; ** p < 0.01; *** p < 0.001; **** p < 0.0001.

The Principal Coordinates Analysis (PCoA) plots indicated the structure of fish communities primarily aligns with their respective streams rather than the period of sample collection, as evidenced by both relative read abundance (RRA) and presence-absence (PA) data (Fig. 4A and B). This result was strongly supported by PERMANOVA analysis (Table 1). Nonetheless, stream-specific variations emerged as the most critical determinant of differences in fish communities, accounting for 38.09% of variations (R²_adjusted), a much higher proportion than that accounted for by the sample collection time (4.70%). Analogous conclusions were clarified using PA data, which revealed an even larger contribution of stream variation (58.64%). Of the streams, Seom displayed the lowest variance in the fish assemblage (median_{centroid−RRA} = 0.37; median_{centroid−PA} = 0.26; Supplementary Fig. S1), whereas the other two streams, Gilan (median_{centroid−RRA} = 0.51; median_{centroid−PA} = 0.28) and Oshipcheon (median_{centroid−RRA} = 0.51; median_{centroid−PA} = 0.42), exhibited higher degrees of dispersion (betadisper, F_RRA = 11.643, p < 0.0001; F_PA = 6.5436, p = 0.0029; Supplementary Table S3). In contrast, no significant variations in fish communities were observed across different mesohabitats, irrespective of the sampling time (pseudo-F_mesohabitat = 1.1133, p = 0.33; Fig. 4C and D, and Table 1). Moreover, the relative count of species recovered did not show significant differences (Kruskall-Wallis, p = 0.68) across the three surveyed mesohabitats, including riffle, run, and pool (Supplementary Fig. S2).

Fig. 4.

Principal Coordinate Analysis (PCoA) of fish communities correspond to month (A, B) and mesohabitat (C, D) as the factor using relative read abundance (A, C) and presence-absence (B, D) data. The shaded ellipse areas show 95% confidence intervals.

Table 1.

PERMANOVA results for overall fish communities using relative read abundance and presence-absence data.

	df	Sums of sqs	Mean Sqs	Pseudo-F	R 2 adjusted	P(perm)
Relative abundance (Bray-Curtis distance)
month	1	0.5963	0.59634	3.6675	0.04707	0.0041
stream	2	5.7285	2.86424	17.6152	0.38095	0.0001
section	2	3.4301	1.71503	10.5475	0.26123	0.0001
mesohabitat	2	0.362	0.18102	1.1133	0.00418	0.33
Residuals	46	7.4796	0.1626
Total	53	17.5966
Presence-absence (Jaccard distance)
month	1	0.4742	0.4742	4.486	0.06064	0.0013
stream	2	8.3047	4.1523	39.284	0.58642	0.0001
section	2	2.0808	1.0404	9.843	0.24671	0.0001
mesohabitat	2	0.171	0.0855	0.809	−0.00712	0.6224
Residuals	46	4.8622	0.1057
Total	53	15.8928

Within a specific stream, distinct community profiles were observed in relation to sampling time and section variability. While Seom showed overlapping communities among sections, the PCoA analysis revealed clear segregation of the upper sections in Gilan and Oshipcheon from the middle and lower sections (Fig. 5A). This pattern was supported by the heatmap analysis of Bray-Curtis and Jaccard distance, characterized by relatively high homogeneity based on RRA (Fig. 5B). Those analyses were supported by a relatively comparable variance explaining of sampling time and section using RRA (23.59% vs. 22.48%, Table 2) and using PA (25.48% vs. 40.32%, Table 2) in Seom. In contrast, Gilan and Oshipcheon exhibited a greater contribution of section variability to the overall variance, with section explaining six and twelve times more variance than sampling time, respectively (Table 2). Notably, Oshipcheon showed the highest variance explained by section variability (R²_{adjusted − RRA} = 72.74; R²_{adjusted PA} = 73.14; Table 2) and displayed a clearly segregated community among the three sections (Fig. 2A and B). An exceptionally high abundance of Rhynchocypris oxycephala (ranging from 32.27% to 87.08%) was observed in the upper sections of Gilan and Oshipcheon (Supplementary Fig. S3 and S4).

Fig. 5.

(A) Principal Coordinate Analysis (PCoA) analysis of fish communities using relative read abundance (RRA) and presence-absence (PA) data in each stream. (B) Heatmap analysis of Bray-Curtis and Jaccard distance from RRA and PA, respectively. The shaded ellipse areas in PCoA show 95% confidence intervals. SM, GN, and OP denotes Seom, Gilan, and Oshipcheon, respectively.

Table 2.

PERMANOVA result of each stream using relative read abundance and presence-absence data presented as the numbers outside and inside parentheses, respectively.

Source of variance	Pseudo-F	R 2 adjusted
Seom
month	8.80 (14.22)	0.2359*** (0.2549***)
section	4.72 (11.46)	0.2248*** (0.4032***)
mesohabitat	0.92 (0.87)	–0.0049 (–0.0051)
Gilan
month	6.20 (13.83)	0.0958** (0.1176***)
section	17.09 (40.54)	0.5922*** (0.7248***)
mesohabitat	0.48 (0.60)	–0.0192 (–0.0073)
Oshipcheon
month	5.74 (6.10)	0.0562* (0.0598**)
section	31.66 (32.23)	0.7274*** (0.7314***)
mesohabitat	1.12 (0.92)	0.0029 (–0.0019)

The analysis was performed using 9999 permutations. Asterisks indicate the significance level (p): * p < 0.05, ** p < 0.01; *** p < 0.001.

Metaphylogeography of fish species

To assess the feasibility of employing eDNA metabarcoding for population genetic analysis of fish species, the obtained amplicon sequence variants (ASVs) from three streams (Supplementary Table S2) were subjected to further analysis. Out of the 76 identified species, 49 exhibited a single ASV, while 27 species had multiple ASV numbers. The three alien species were excluded from analysis, including Cyprinus carpio, Micropterus salmoides, and Oncorhynchus mykiss. Among the species with multiple ASVs, Coreoperca herzi (Herzenstein, 1896) and Rhynchocypris oxycephala had the highest number of ASVs (n = 4). Four species, including Microphysogobio yaluensis (Mori, 1928), Odontobutis interrupta (Iwata & Jeon, 1985), Misgurnus anguillicaudatus, and Phoxinus phoxinus, had three ASVs. Notably, the highest number of segregating sites was observed in Coreoperca herzi and Misgurnus anguillicaudatus (n = 13), followed by Microphysogobio yaluensis and Zacco platypus (n = 6), and Pungtungia herzi (n = 5) (Fig. 6). Geographical ASV distributions were clearly identified in three species, including Odontobutis platycephala (Iwata & Jeon, 1985), Coreoperca herzi, and Microphysogobio yaluensis. These species exhibited Seom- and Gilan-specific haplotypes, representing the WKS and SKS, respectively. No Oshipcheon-specific haplotypes were observed (Fig. 6). In contrast, the remaining fish taxa displayed mixed populations across different geographic regions. By conducting ASV network analysis and comparing historical translocated native species among streams², we identified the potential sources of eight translocated species in east sea flowing streams from west and/or south sea flowing streams. These species include previously reported taxa such as Pungtungia herzi, Coreoleuciscus splendidus, Nipponocypris koreanus, Zacco platypus, Koreocobitis rotundicaudata (Wakiya & Mori, 1929), and Odontobutis interrupta (Table 3, Supplementary Table S4). However, the ASV network analysis did not reveal any genetic evidence of species translocation to the west and south sea flowing streams from other regions.

Fig. 6.

ASV networks of several fish species correspond to their geographic regions inferred from MiFish metabarcoding studies of three South Korean streams. Each circle designates an identical ASV and the size is proportional to the semiquantitative percentile of respective ASVs. The small, black circles denote hypothetical intermediate ASV to connect obtained ASVs by this study. Each hatch mark indicates a single mutation step among ASVs, disregarding its line length. Asterix showed putative translocated ASV to Oshipcheon.

Table 3.

Fish species and possible sources of putative translocated species to Oshipcheon.

Species	Native habitat a	Putative source(s)
Coreoleuciscus splendidus	Han and Nakdong River	WKS
Koreocobitis rotundicaudata	Han River	WKS
Nipponocypris koreanus	All streams but NKS	SKS/WKS
Odontobutis interrupta	Han and Geum River	WKS
Pungtungia herzi	All but NKS	SKS & WKS
Zacco platypus	All streams but NKS	SKS/WKS
Misgurnus anguillicaudatus	All	Imported
Rhynchocypris oxycephala	N/A	N/A

^a Original data source from Yoon, Kim². the Northeast Korea Subdistrict (NKS), the South Korea Subdistrict (SKS), and the West Korea Subdistrict (WKS).

Discussion

We successfully obtained a comprehensive dataset of 107 fish amplicon sequence variants (ASVs) from 54 water samples collected across three small streams in South Korea. These ASVs spanned ten orders, covering 20 families, 46 genera, and 76 species, representing approximately one-third of Korea’s total freshwater fish species (220 species). The remarkable diversity captured from just nine sampling sites in small streams highlights the efficiency of eDNA metabarcoding compared to traditional survey methods. The approach demonstrated superior performance, recovering significantly more species, with a substantial overlap of 72.22%, 80%, and 86.49% of species previously reported in Gilan¹⁷, Oshipcheon¹⁸, and Seom¹⁹, respectively. Furthermore, eDNA metabarcoding detected an additional 10 to 17 species each stream not identified in traditional surveys, showcasing its sensitivity. Some species detected exclusively in traditional methods were rarely observed, potentially due to methodological biases²⁰ or were recognized as having low abundance in traditional survey contexts. As sampling efforts increase, the likelihood of detecting rare species also rises. Overall, eDNA metabarcoding offers a powerful tool for large-scale, long-term monitoring of fish communities, providing high detection sensitivity, minimal bias, and low environmental impact.

eDNA metabarcoding effectively revealed distinct biogeographical patterns of freshwater fishes across the Korean Peninsula, demonstrating strong congruence with known zoogeographical boundaries and endemicity patterns. The analysis showed that fish assemblages were primarily structured by the presence of endemic species and the unique environmental characteristics of each stream (Fig. 2). PERMANOVA and heatmap analyses confirmed significant relationships between fish community composition and stream variations. Endemism, often higher in river tributaries⁹, was notable in our study, ranging from 37.71% in Oshipcheon to 40.7% in Seom, aligning with or exceeding prior surveys^2,15,17,19. eDNA metabarcoding approach consistently detected endemic species, producing results comparable to previous traditional surveys. Heatmap clustering further emphasized the influence of endemic species, such as Rhodeus pseudosericeus, Gobiobotia brevibarba, and Gobiobotia macrocephala in Seom, and Squalidus multimaculatus in Oshipcheon^19,21–23. Conversely, Acheilognathus majusculus, Koreocobitis naktongensis, Niwaella multifasciata, and Tanakia latimarginata were limited to the SKS region and are known endemic species of the Nakdong riverine system^11,24,25. The NKS (Oshipcheon) was clearly distinct from the other biogeographical regions. In contrast, the SKS (Gilan) was closely related to the WKS (Seom), supported by higher shared species between Gilan and Seom. Similarly, the pattern of shared species among three geographic distribution of freshwater fishes was previously observed in the streams of national parks²⁶, indicating high similarities of fish communities between SKS and WKS. Therefore, eDNA metabarcoding is a valuable method for unrevealing biogeographical pattern as reported in coastal ecosystems²⁷. By focusing on tributaries and small streams, our study revealed a high prevalence of endemic species and distinct fish assemblages in these habitats, underscoring the utility of eDNA metabarcoding for assessing biodiversity in tributary or highland streams and revealing biogeographical patterns of freshwater fishes.

Not only reflected by the presence of endemic species, fish assemblages within individual streams mirrored the specific environmental characters of each section. The upstream-specific species Rhynchocypris oxycephala was notably high in the upper reaches of both Gilan and Oshipcheon streams, whereas its presence was not particularly abundant in the upper section of the Seom stream (Supplementary Fig. S3 and S4). This species, known to favor cold-headwater streams^28,29, is found most abundantly at the highest elevations in the Korean peninsula, underlining its key role in the upstream of these riverine systems^26,30,31. Furthermore, the lower-reach inhabiting species (Cottus hangiongensis, Gymnogobius petschiliensis (Rendahl, 1924), or Squalidus multimaculatus) and diadromous species (Oncorhynchus keta and Tribolodon hakonensis) were the designated species in lower section of Oshipcheon (Supplementary Fig. S4). Although metabarcoding data can be biased with respect to species abundance³², the method has shown strong congruence with traditional surveys and is generally robust for assessing fish community composition³³. Furthermore, our study employed both read abundance and presence–absence datasets to characterize freshwater fish communities, revealing consistent patterns across analyses (Table 2).

Although variations in physicochemical parameters and food web structures can influence fish communities across mesohabitats³⁴, our study found no significant differences among riffle, run, and pool habitats within the same site (Supplementary Fig. S2). In small streams or tributaries, eDNA demonstrated homogeneity due to lateral dispersion, forming a plume distribution with downstream heterogeneity. Lateral dispersion refers to the horizontal spreading of eDNA across the stream channel caused by turbulence and hydrodynamic mixing, which results in eDNA being distributed not only downstream but also toward the banks^35,36. In contrast, slower water velocities in larger rivers or lakes may amplify mesohabitat influences on eDNA³⁷. Tailored sampling strategies are vital to optimize eDNA metabarcoding in diverse aquatic systems.

eDNA analysis not only revealed the spatial and biogeographical distribution of freshwater fishes in Korean streams, particularly in Oshipcheon, but also demonstrated their temporal dynamics. This study highlighted seasonal variations in fish assemblages, reflecting changes in species richness and community composition over time. These patterns likely correspond to the life cycles of various fish species inhabiting the stream^38,39. Despite their small catchment area⁴⁰, streams in the NKS are important ecosystems for diadromous fishes. We observed the eDNA dynamic of migratory fishes in Oshipcheon corresponding to their seasonal movements. In March, a low abundance of Oncorhynchus keta was observed in the lower Oshipcheon, while this anadromous salmon was not detected in August, indicating that the fry remained in the stream until April before migrating to the ocean⁴¹. Conversely, the spring spawner T. hakonensis was detected in both sampling times, but its RRA was higher in March compared to August, reflecting the migration event of this anadromous minnow in the Korean east coastal area⁴². Furthermore, the amphidromous Plecoglossus altivelis was exclusively detected in August in the mid-lower Oshipcheon, with no evidence of their eDNA observed in March. This aligns with a similar migratory model of eDNA detection exhibited by species-specific assays⁴³. The observed dynamics of RRA corresponding to the seasonal movement of migratory fishes have been previously reported in urban estuaries⁴⁴ and tributary³⁸. Furthermore, species-specific eDNA studies have confirmed the spatiotemporal eDNA concentration of migratory fishes^45–47. The findings of this study highlight the benefit of eDNA metabarcoding to reveal seasonal variations on fish assemblages.

Seasonal variation appeared to contribute more significantly to the Seom communities than to the other two streams, which is likely related to the stable depth typical of a lowland stream like Seom. The Seom, characterized by lower altitude, milder gradients, and consistent water depth across its sections (Supplementary Fig. S6), exhibited the highest species richness (n = 49), a typical feature of lowland streams⁹. These stream characteristics facilitate fish migration and amplify seasonal influences on fish communities. Additionally, fish assemblages in Seom showed high homogeneity across section compared to Gilan and Oshipcheon, aligning with the characteristics of lowland streams in the Han riverine system and supporting findings from previous research^48,49.

Beside identifying the fish assemblage, the ASVs were further scrutinized to juxtapose haplotypes across three biogeographical subdistricts in South Korea. Advanced bioinformatics tools, including denoising and abundance curation, allowed ASVs to approximate true haplotypes⁵⁰, supporting studies of metaphylogeography. While several metaphylogeographic studies have been conducted on marine organisms^51,52 or arthropods^53,54, only a few have focused on freshwater fishes, and these were generally restricted to specific genera or species^55,56. Current study identified multiple haplotypes corresponding to various freshwater fish species.

Distinct ASVs of Coreoperca herzi, Odontobutis platycephala, and Microphysogobio yaluensis were observed between Seom and Gilan, indicating an allopatric population between WKS and SKS. Concordance to our result demonstrating 0.6% to 8.6% inter-haplotypic Kimura 2-parameter (K2P) distance, a previous phylogenetic analysis also showed a higher divergence of Coreoperca herzi populations in South Korea, indicating a clear separation between the WKS and SKS populations using cytochrome oxidase I (COI) marker⁵⁷. Nonetheless, the phylogeographical analysis of the allopatric populations of Microphysogobio yaluensis and Odontobutis platycephala has been insufficiently documented, suggesting that the population genetics of these two endemic species necessitate considerable management efforts. The phylogeography of Nipponocypris koreanus and Coreoleuciscus splendidus aligns with earlier population genetic investigations utilizing COI¹ and mitogenome⁵⁸, respectively, demonstrating a contrasting divergence between WKS and SKS. Moreover, the phylogeographical studies of freshwater fishes within the Korean Peninsula are extensively documented, revealing significant genetic divergence among populations of various native species from distinct biogeographical regions, such as Rhodeus notatus (Nichols, 1929)⁵⁹, Rhodeus pseudosericeus²¹, Rhodeus kumgangensis⁶⁰, Barbatula nuda (Bleeker, 1864)⁶¹, and Cobitis nalbanti (Vasil’eva, 2016)⁶², and Oryzias latipes (Temminck & Schlegel, 1846)⁶³. Earlier phylogeographical studies have proposed that the Paleo-Yellow River, Baekdudaegan Mountain Ranges, and Okcheon Belt served as significant barriers, effectively partitioning the Korean Peninsula into three principal biogeographic regions, which consequently facilitated allopatric separation. This segregation impeded gene flow, subsequently facilitating the speciation of freshwater fishes^58,64.

Together to Coreoperca herzi, Misgurnus anguillicaudatus exhibited the highest segregating sites (n = 13), reflecting inter-haplotypic pairwise K2P distance spanning between 3% and 7.25%. While genetic studies of Misgurnus anguillicaudatus in Korean waters are notably sparse, this particular loach has shown significant nucleotide divergence in Japan (0.3% to 4.0%) and Mainland China (0.1% to 7.3%), suggesting the existence of at least two clades within each country^65–67. Furthermore, Rhynchocypris oxycephala also harbored considerable sequence divergences, with nine segregating sites, displaying two ASVs (ASV_021 and 108) closely associated with Rhynchocypris lagowskii. Correspondingly, multiple haplotypes of Rhynchocypris oxycephala and Rhynchocypris lagowskii from the northern Far East exhibited low divergence, suggesting mitochondrial introgression due to historical hybridization events⁶⁸.

Mitochondrial haplotype analyses have been utilized for tracing the potential origin and dispersal of introduced fish species within a specific region^69–72. Our investigation identified five probable species translocations, including Nipponocypris koreanus, Pungtungia herzi, Zacco platypus, Misgurnus anguillicaudatus, and Rhynchocypris oxycephala (Fig. 6). According to a national survey conducted in 2014, a total of 28 translocated species have been recorded in Korea². All putative translocated species detected in our research coincide with the survey list, proposing that eDNA metabarcoding could serve as a valuable tool for monitoring genetic disturbances caused by various anthropogenic activities. Furthermore, the origins of the three species with mixed haplotypes (Coreoleuciscus splendidus, Koreocobitis rotundicaudata, and Odontobutis interrupta) were unclear, it was ambiguous whether they were introduced species² or naturally emerged prior to the WKS-SKS separation due to the Okcheon Belt formation during the early Triassic to Jurassic periods⁷³, predating the eastward uplifting in the late Oligocene⁷⁴. Additionally, the reduced sea levels during the Last Glacial Maximum led to the connection of several basins⁶⁴, facilitating more complex speciation events within the Korean Peninsula.

Interestingly, two ASVs of Koreocobitis rotundicaudata consisted of a unique NKS (ASV_071) and a shared ASV of SKS and NKS (ASV_028), demonstrating a distinct population due to the Late Pleistocene divergence¹⁰ and intentional release to reestablish their population in NKS streams^2,11, respectively. Surprisingly, a pattern akin to the introduced Coreoleuciscus splendidus population in Oshipcheon and the native WKS and SKS population was observed by previous work using 16 S marker⁷⁵, which was corroborated by ASV network in this study. Our results indicated that the Nipponocypris koreanus population in Oshipcheon originated from WKS and SKS, however, a preceding phylogeographical study¹ proposed that this species was artificially introduced to NKS from WKS, evidenced by a minimal average genetic distance (0.4%) of the monophyletic clade between them. Still, there is a lack of evidence for a mixed population with SKS presented in the reference study, which could be due to a smaller translocated population of SKS individuals in Oshipcheon, as presented in our study. When excluding the ASVs of the six translocated fishes in Oshipcheon, we observed phylogenetically different populations of Coreoleuciscus splendidus and Nipponocypris koreanus between Seom and Gilan, and K. rotindicaudata between Seom and Oshipcheon, highlighting phylogeographical patterns of freshwater fishes in the Korean Peninsula. Additionally, it is worth noting that eDNA metabarcoding using the universal MiFish primer may not provide sufficient sequence variability to discriminate populations between Seom and Oshipcheon. Overall, our metaphylogeography study presented a promising approach for tracking the original sources of introduced populations and revealing phylogeographic patterns using multispecies eDNA metabarcoding from water samples, supporting tissue-based phylogeographic studies of freshwater fishes in the Korean Peninsula.

While eDNA haplotyping has shown promise, it remains premature to replace traditional population genetic studies due to uncertainties in eDNA metabarcoding quantification³². Improvements, including the incorporating quantified internal positive controls in high-throughput sequencing⁵⁶, offered a progress, but validating eDNA read numbers as proxies for population metrics remains essential. Our study showed a low number of haplotypes per species, which might be due to the target sequence being shorter and fewer variable sites compared to markers like ND2 or COI⁷⁶. Current metabarcoding techniques usually use short mitochondrial ribosomal markers, which do not allow for high resolution of haplotypes. The extended sequence of the Fish-U primer, spanning from 12 S to 16 S rRNA⁷⁷, may improve haplotype acquisition. However, its reliability for eDNA samples needs to be verified, since long sequences may reduce recovery rates^78,79 and limited number of available databases should be considered.

Conclusion

Our study highlights the potential of eDNA metabarcoding derived from water samples to elucidate not only the geographical distribution but also the phylogeographical patterns of endemic freshwater fishes communities within South Korea’s biogeographical regions. This can be achieved through surveys of tributaries or highland streams, thereby mirroring the spatiotemporal dynamics obtained from traditional fish surveys. Analysis of their ASV networks could also help infer the origin(s) of recorded species translocations. Furthermore, we emphasize the importance of considering the field sampling design, specifically with respect to the timing of sampling and the mesohabitat, to optimize fish detection. These insights could substantially advance metaphylogeography studies using eDNA and multispecies markers, and provide crucial contributions to the formulation of appropriate field sampling strategies for eDNA surveys of freshwater fishes in wadable streams.

Materials and methods

Study Sites, field Sampling, and water collection

Three streams, including Seom, Gilan, and Oshipcheon, were selected to represent the primary biogeographical distributions of freshwater fishes in South Korea (Fig. 7). Each stream was further divided into three sites (upper, middle, lower), and the mesohabitat of each site (riffle, run, and pool) was determined through visual identification⁸⁰. Water samples of one liter each were collected from the surface of the streams during the spring (March) and summer (August) of 2020. Water temperature and salinity were measured using the HI-98,194 multiparameter meter (Hanna Instruments Inc., USA), while flow rates and depth were assessed using the FP111 water velocity meter (Global Water Flow Probe, Texas, USA). To ensure the removal of potential contaminants, all equipment utilized in the sampling process were decontaminated using a 10% v/v commercial bleach solution containing sodium hypochlorite (7.5%) for 10 min. Immediately after collection, the water samples were preserved on ice and transported to the laboratory for filtration. A vacuum filtration system was employed through 0.45 μm GN-6 Metricel membranes (PALL Life Sciences, USA), with each one liter water sample divided equally and filtered using two membranes (500 mL per filter). Additionally, a sterile deionized water-filtered membrane served as the blank control in the experiment.

Fig. 7.

Nine study sites among the three streams (Gilan, Seom, and Oshipcheon). The solid red line showed the mountainous terrestrial barriers for generating three main biogeographical subdistricts in South Korea. U, M, and L denotes the upper, middle, and lower section, respectively. The map of the Korean Peninsula was generated using MATLAB R2021a, and the individual stream maps were produced using rgdal and ggplot2 package in R v3.6.3.

Environmental DNA extraction and metabarcoding

DNA was extracted from each membrane using the DNeasy Blood & Tissue Kits (Qiagen GmBH, Hilden, Germany) following the manufacturer’s instructions. In summary, the membrane was placed in 2 mL tube containing 630 µL of ATL buffer and ceramic spheres (6.35 mm diameter), then disrupted using the FastPrep-24™ Classic Instrument (MP Biomedicals, Irvine, CA, USA). The extracted DNA were aliquoted and stored at − 20 °C until further analysis using metabarcoding.

For high-throughput sequencing, a library was prepared by PCR amplification using the MiFish universal primer⁸¹. The PCR mixture (20 µL) contained 2.0 µL of the extracted DNA from a subsample, 1.0 µL each of the forward and reverse primers, 2.0 µL dNTPs (2.5mM), 2.0 µL buffer, 0.2 µL ExTaq Hot Start polymerase (Takara Bio Inc., Tokyo, Japan), 0.6 µL DMSO (3%), and 11.20 µL of sterilized deionized water⁸². The PCR conditions included an initial denaturation step at 95 °C for 3 min, followed by 35 cycles of 94 °C for 20 s, 60 °C for 15 s, and 72 °C for 15 s, with a final extension at 72 °C for 5 min. The PCR products of the expected size (~ 300 bp) were purified using the AccuPrep^® Gel Purification Kit (Bioneer, Republic of Korea) and subsequently used as a template for the second indexing PCR with the Nextera XT index kit (Illumina, San Diego, CA, USA). The second amplicons (~ 350 bp) were quantified using the Qubit dsDNA HS Assay Kit (Invitrogen, USA) and sequenced using the Illumina MiSeq platform (2 × 300 bp).

Bioinformatic processing and data analysis

The primers were removed using Cutadapt v2.10⁸³ after adapter and low-quality (QV < 20) trimming by CLC Genomics Workbench v8.0 (CLC Bio, Cambridge, MA, USA). Following overall quality assessment, the primer-free sequences were denoised using Divisive Amplicon Denoising Algorithm 2 (DADA2) v.1.16.1⁸⁴ with specific parameter adjustments, including maxEE = c(1,1)⁸⁵ and truncLen = c(120,100). Decontamination of amplicon sequence variants (ASVs) was performed by implementing the prevalent blank correction method from the decontam package v1.6.0⁸⁶. To minimize false-positive detections, ASVs with low abundance (< 0.1% per library) or low occurrence (single detection in a river) were excluded from further analysis⁸⁷. The abundance-based post-denoising algorithm was applied using LULU v 0.1.0⁸⁸ with a minimum match parameter set at 95 to obtain curated ASVs representing fish haplotypes.

Prior to taxonomy assignment by BLAST v2.10.1+⁸⁹, a custom fish database⁹⁰ was generated by downloading mitochondrial sequences of non-tetrapod vertebrates from GenBank (https://www.ncbi.nlm.nih.gov/genbank/, accession date: 20/012/2022) and constructing the database using Pipecraft 1.0⁹¹. Curated ASVs with ≥ 97% identity were assigned the top-scored species name, those with 95%≥ identity > 97% were assigned the top-scored genus without species identity, and ASVs with < 95% identity were classified as “unknown”^77,92. The BLAST results were validated by aligning ASVs with the queried NCBI databases using MAFFT⁹³, and phylogenetic construction was conducted using PhyML⁹⁴. Non-fish and “unknown” sequences were excluded from the community analysis.

The evaluation of fish community biodiversity metrics, including alpha and beta diversity, based on RRA^95,96 and presence-absence (PA)³⁹, was performed using phyloseq v1.32⁹⁷ and vegan v2.5.7⁹⁸. Principal Coordinate Analysis (PCoA) was conducted using the ordinate function to visualize the fish community structures, followed by permutational multivariate analysis of variance (PERMANOVA)⁹⁹ to assess statistical differences. The reliability of the PERMANOVA results was tested using the betadisper function. Heatmap analysis was contructed using the pheatmap v1.0.12 package¹⁰⁰ by combining the dataset of each mesohabitat in the respective stream sections and transforming it into presence-absence (PA) data. Bray-Curtis distance metric was implemented for all distance-based permutational analyses related to beta diversity. The visualization of shared species among streams was generated using the ComplexUpset v1.3.1 package^101,102. Venn diagrams were created using the ggVennDiagram v1.2.1 package¹⁰³ to compare the species detected by eDNA analysis and the existing fish survey data for each stream. All analyses were performed using R v3.6.3 (http://www.r-project.org) and the corresponding packages as mentioned.

The genetic distances between aligned amplicon sequence variants (ASVs) were calculated using the K2P model¹⁰⁴ implemented in MEGA-X¹⁰⁵. TCS networks¹⁰⁶ were generated using PopART v1.7 software¹⁰⁷, and the semiquantitative percentile ranking (0–4) of each ASV was employed as a representation of haplotype abundance^7,108. The networks were constructed for species that possessed more than a single ASV and exhibited informative phylogeographical features.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

M.H.F.A., A.R.K, H.J.L., and H.W.K. conceptualized the study; H.W.K., M.H.F.A., A.R.K, H.W.B., and J.E.J. performed the research methodology; J.E.J., M.H.F.A., and A.R.K contributed to bioinformatic pipeline for study; M.H.F.A., A.R.K, H.W.B., J.E.J., H.J.L. and H.W.K. conducted formal analysis and data curation; H.W.B., H.J.L. and H.W.K. supervised the study; and M.H.F.A., A.R.K, and H.W.K. wrote the original draft. All authors contributed to the review, writing, and editing of the final draft.

Funding

This work was supported by a grant from the National Institute of Biological Resources (NIBR), funded by the Ministry of Environment (MOE) of the Republic of Korea (NIBR201933203), and partially supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (RS-2021-NR060118).

Data availability

The raw sequences of the samples are available on the Sequence Read Archive (SRA) under the name Bioproject PRJNA994282.

Declarations

Competing interests

The authors declare no competing interests.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.