Large-scale marine eDNA database reveals seasonal latitudinal diversity gradients in Japan driven by vagrant fish

Understanding the spatial and temporal distribution patterns of marine biodiversity is becoming increasingly important for the management and sustainable use of fishery resources, as well as for the conservation of marine ecosystems. Fish diversity, which contributes to ecosystem functioning, is critical to the provision of all ecosystem services^{1, 2}. However, assessing systematic patterns of biodiversity remains a key challenge, especially in the context of biodiversity loss³, as the marine ecosystem is inherently variable and strongly influenced by environmental changes and fishing pressure at global scales.

Although ocean covers approximately 70% of the Earth’s surface and harbors tremendous biodiversity potential⁴, its vastness and inaccessibility make it difficult to access the actual state of biodiversity loss compared to terrestrial ecosystems. Hence, it is critical to develop accurate and spatially extensive monitoring methods to understand current patterns of fish diversity.

Efforts to evaluate global-scale marine biodiversity patterns have advanced by initiatives such as the Census of Marine Life (CoML, http://www.coml.org/), which has generated valuable resources for identifying and documenting marine species (Cosello et al., 2010⁵. Yet, capturing biodiversity patterns require accounting for spatiotemporal variations. Seasonal migratory fish alter biodiversity patterns over time, appearing in different area depending on the season^{6, 7, 8, 9}. In addition to migration, the temporal appearance of vagrant fish, those that move beyond their native range but fail to establish persistent populations, can also influence seasonal changes in biodiversity^{10, 11, 12}. Spatial variation, meanwhile, is shaped by ocean currents and the structure of the littoral zone¹³. Therefore, both temporal and spatial context must be considered to assess correctly the marine fish biodiversity. However, there is limited spatiotemporal biodiversity information available using a consistent method, hampering our ability to assess the recent impacts of human activities and climate change.

Large-scale monitoring of fish diversity is constrained by high logistical and financial demands^{4, 14}. A promising solution is environmental DNA (eDNA), which refers to DNA shed by organisms into their environment^{15, 16}. Recent advances in eDNA analysis have made it a reliable, noninvasive, and highly sensitive monitoring tool for estimating diversity and species composition, especially in aquatic ecosystems^{9, 17, 18, 19}. Compared with traditional capture-based surveys, eDNA offers relatively lower cost and higher scalability, allowing for more frequent and extensive spatio-temporal sampling²⁰. Because eDNA can originate from a wide variety of sources, including intestinal and skin cells, urine, mucus, eggs, or sperm²¹, it enables detection of multiple species from a single water sample^{22, 23}.

To investigate the spatiotemporal changes in fish diversity patterns in the marine system, we utilized data from the All Nippon eDNA Monitoring Network (ANEMONE, https://db.anemone.bio), a publicly accessible nation-wide eDNA-based biodiversity monitoring program in Japan. Established in 2019, ANEMONE currently hosts more than 8,000 surveys with data continuously updated on the database (Figure. 1²⁴. The database covers over 3000 km²of coastal marine areas, including regions influenced by the Kuroshio and Oyashio currents, two major western boundary currents in the western North Pacific. As of September 2024, the database includes data from 2017 to 2022. Sampling sites are mainly located along the coastline of Japan, – an area of high biodiversity and ecosystem service value due to the intersection of the land and sea and the complex mosaic of topographic and environmental conditions¹³. Japan harbors at least 3,790 fish species, reflecting a combination of various historical and environmental factors¹⁴. As a major fishing nation, Japan also holds potential to sustainably enhance its fishery yields through appropriate management^{25, 26}. Thus, the ANEMONE provides an ideal basis for analyzing spatial and temporal patterns of marine fish diversity. These data are publicly available as the ANEMONE database (https://db.anemone.bio), and this study utilized this database for analysis.

To explore large-scale patterns of marine biodiversity, we focused on the inter-regional and inter-seasonal differences in the latitudinal diversity gradients (LDGs). The LDG is one of the most well-established global ecosystem patterns, generally showing that species richness increases toward lower latitudes, particularly in terrestrial systems^{27, 28}. Similar patterns have often been reported for marine taxa in both coastal and deep-sea ecosystems^{27, 29, 30, 31, 32, 33}, although there are several exceptions^{34, 35}. Despite extensive studies on LDGs, little is known about how they vary temporally, especially across seasons. Fish, being highly mobile, migrate in response to changes in environmental conditions. A prominent seasonal pattern is the northward migration in summer and return to southern waters in winter (e.g., Furukawa et al.³⁶,, which is expected to flatten or even reverse the LDGs in winter as more fish concentrate at higher latitudes. Alternatively, the seasonal influx of vagrant fish may also influence the LDGs. For example, many tropical fishes are observed in Japanese waters in summer, brought by the Kuroshio Current, but most do not survive the winter^{10, 11, 12}. The transient increase in tropical species richness at higher latitudes may steepen the LDGs in summer, strengthening the expected steeper slope arising from the seasonal fish migration patterns. Given those dynamics, we investigated how LDGs of marine fish diversity change seasonally in Japanese waters. The Japanese Archipelago spans approximately 3,000 km from north to south and encompass a range of climate zones from subboreal to subtropical^{14, 25}, making it an ideal system to study LDGs.

In this study, we aimed to (1) characterize the general LDGs of marine fish in Japanese waters, (2) examine the seasonal variations in LDGs, and (3) explore the potential drivers of the seasonal changes in the LDGs (i.e., the effect of seasonal migration and temporary inflow of fish), focusing on the roles of vagrant and migration fish.

To assess the inter-regional fish LDGs, Japanese waters were divided into two regions: the Mainland and the Ryukyu area (Fig. 1). In the raw data (i.e., before rarefaction), the mean genus richness value in the Mainland area was 25.33 for 1,119 surveys (median = 23, range 4–78, mean = 25.33, SD = 11.58) and the value in the Ryukyu area was 41.52 for 155 surveys (median = 40, range 10–90, mean = 41.52, SD = 15.31). The number of surveys obtained by season and latitude is summarized in Table S1. Latitude was negatively correlated with rarefied fish genus richness in the Mainland area (Fig. 2a; Table 1), whereas significant patterns were not observed in the Ryukyu area (Fig. 2b; Table 1). In the models, the coefficients in the Ryukyu area differed from those in the Mainland area. In spring, summer and autumn, the Mainland area exhibits significantly higher genus richness compared to the winter season. The interaction between seasonal variation and latitude was significant for the Mainland area (Fig. 1a; Table 1). Remarkably, the interaction between spring, summer and autumn and latitude was significant. The slope of the regression line for winter was less steep than that for the other warm seasons (Fig. 2a). In other words, during the warm seasons genus richness is significantly higher in high-latitude regions. Note that, we found the similar negative LDGs pattern for the data pooled for all sampling surveys, but the interaction between season and latitude was not observed from the data (Figure S1, Table S2).

To identify the factors driving the seasonal differences in the LDGs, we investigated two potential mechanisms. The first was presence of the season-specific fish, defined as genera observed exclusively only in one season at a given sampling site (e.g., a species present in summer but absent in the other seasons at the site). For each season, we calculated the proportion of survey sites in mainland Japan (i.e. excluding Ryukyu area) that harbored more than one season-specific genus.

We found that the occurrence of season-specific genera was highest in summer with 16.92%, compared to 7.35% in spring, 10.33% in autumn, 4.69% in winter (Fig. 3). A one-sided binomial test, using the overall data all seasons as the reference, revealed that the proportion of season-specific fish was significantly larger in summer (p = 0.976 in spring, p < 0.001 in summer, p = 0.647 in autumn, p = 1.000 in winter). Although the difference was not statistically significant, the values were higher in spring and autumn than in winter. In addition, those warm season-specific fishes tended to occur at lower latitudes (Fig. 3b), suggesting that many of them were tropical vagrant fish. Additional analyses using stricter criteria, requiring repeated detection within a season, yielded qualitatively similar patterns, suggesting that the results were not solely driven by rare or sporadic detections.

The second factor examined was the contribution of migratory fish. We visualized the seasonal migration patterns of five genera of fish that are designated as useful fish in resource management by Japanese Fisheries Agency – Cololabis, Sardinops, Scomber, Seriola and Trachurus. However, the spatial distribution of these genera did not exhibit clear or consistent seasonal shifts across the study period (Fig. 4), suggesting that seasonal migration by these taxa may have limited influence on the observed seasonal variation in the LDGs.

To determine whether the results were due to fish movement, we applied the same GLM method to inland waters where the sampling sites were separated. We found no seasonal change in the LDG slope in inland waters (Figure S2, Table S3) while the negative LDG patterns were observed similarly to coastal waters. Notably, spring showed a significant deviation from other seasons in inland systems (Table S3).

We observed a positive correlation between water temperature and genus richness (Fig. 5a). The analysis was conducted on the 929 surveys where temperature data existed. The correlation coefficient was 0.407 (p < 0.001). The water temperature strongly correlated latitude (Fig. 5b). The correlation coefficient was − 0.422 (p < 0.001). Thus, we did not include water temperature as an explanatory variable for the original model to avoid the multicollinearity, but we obtained a similar result from the model including water temperature (Table S4).

This study demonstrates that large-scale eDNA observation from the ANEMONE database capture a clear latitudinal diversity gradient (LDG) in marine fish communities along the Japanese coastline. In the Mainland area, fish genus richness declines with latitude, consistent with well-established, globally observed macroecological patterns^{27, 37}. When all sampling data are pooled, a similar declining trend is observed across Japanese waters, supporting the existence of an LDG in Japanese waters at the genus level^{8, 38}.

A key finding of this study is that the strength of LDGs vary seasonally. The interaction between the latitude and season in the Mainland area indicates that LDGs become steeper (more negative) during the warm season relative to winter. Importantly, this pattern should be interpreted as a contrast between winter and the warm seasons collectively, rather than as a phenomenon unique to summer alone. Although the statistical signal is strongest in summer, similar trends were also observed in spring and autumn, suggesting a common underlying mechanism operating across the warm period.

In contrast, no clear LDG pattern was detected in the Ryukyu area. Several non-exclusive explanations may account for this result. First, tropical regions generally exhibit weaker seasonality than to temperate regions, and many tropical species show limited seasonal migration or turnover, which may reduce detectability of seasonal or latitudinal patterns. Second, the latitudinal extent of the Ryukyu area is considerably narrower than that of the Mainland area (approximately 4° vs. 15°), which may limit the ability to detect broad-scale gradients. In addition, uneven spatial sampling may further obscure such patterns.

We evaluated multiple potential explanations for the observed seasonal restructuring of LDG patterns. First, if the seasonal migration were dominant driver, one would expect a flatter LDG in summer due to northward movements of fish populations. However, the seasonal distribution of five migratory genera (Cololabis, Sardinops, Scomber, Seriola and Trachurus) showed little seasonal change, suggesting that migration alone does not explain the observed pattern. Second, temperature-dependent variation in eDNA detectability could potentially inflate apparent richness in warmer seasons. While this mechanism cannot be fully excluded, the absence of comparable seasonal changes in inland waters—where oceanic transport processes are minimal—suggests that detection bias alone is unlikely to account for the observed patterns.

Instead, our results are most consistent with the contribution of transient taxa, particularly tropical vagrant fish transported by ocean currents that fail to establish populations due to low winter temperature^{39, 40, 41}. In Japan, such vagrant fishes are known to be advected poleward by the Kuroshio Current from summer through fall and decline during winter^{42[12, 38, 43]10[,11[,8, 12} where their presence can increase observed diversity⁴⁴. However, because season-specific genera in this study are defined operationally at the regional scale, it is difficult to directly verify the origin and ecological status of each taxon. In addition, species within the same genus may exhibit different seasonal or biogeographic patterns, and taxon-by-taxon validation was not feasible for this large-scale dataset. Moreover, the Ryukyu area, which represents a source area of many tropical fish, may show weaker seasonal patterns. Therefore, while the observed spatiotemporal patterns are consistent with the vagrant fish hypothesis, further studies will be required to directly confirm their contribution to seasonal variation in LDG.

These findings highlight that biodiversity patterns inferred from eDNA-based richness metrics can reflect not only stable community structure but also transient ecological processes. In particular, seasonal increases in observed richness during the warm period may partly represent a temporary inflation driven by non-resident taxa. This has important implications for biodiversity monitoring and assessment, especially under ongoing climate warming, which may further enhance the frequency and extent of such transient occurrences.

At the same time, several limitations should be acknowledged. eDNA signals can persist and be transported, and detection probabilities may vary with environmental conditions such as temperature and water movement. Importantly, these limitations apply to all detected taxa, not only to migratory species, and should be considered when interpreting eDNA-based diversity patterns. In addition, spatial autocorrelation and uneven sampling effort may influence the results, although the inclusion of inland comparisons and rarefaction procedures partially mitigate these concerns. Furthermore, because this study relies on genus-level identification, some ecological signals may be masked or aggregated. Nevertheless, supplementary species-level analyses, where reliable taxonomic assignment was available, showed qualitatively similar patterns, supporting the robustness of our conclusions with respect to taxonomic resolution. Future studies integrating species-level resolution, independent observational data, and mechanistic approaches will be essential to fully resolve the processes underlying seasonal biodiversity dynamics.

Overall, our results demonstrate that large-scale, standardized eDNA monitoring can reveal not only broad spatial gradients but also their temporal variability. The standardized observation framework of the ANEMONE database enables consistent large-scale analyses, highlighting the importance of harmonized eDNA datasets for detecting ecological patterns. Understanding such dynamic patterns will be essential for interpreting biodiversity change and for developing robust indicators in a rapidly changing marine environment.

Data collection

Data sets were obtained from the ANEMONE database (downloaded on 20th September, 2024). The database contains data on both marine and inland water, although we focused on analyzing data on marine fish. To exclude inland water data, we utilized information on coastlines from Nature earth (https://www.naturalearthdata.com), and we have screened the surveys within 10 km of the coastline. The data sampling sites were distributed between 24.46 N and 45.54 N and 123.8E to 145.6E, covering coastal areas throughout Japan. The data collection period was from December, 2017 to September, 2022. The total number of surveys for the five-year time-period was 1274. The raw data from the ANEMONE database consisted of sequence data, species information from the sequence data, and sample metadata including survey location and sampling date. The eDNA data were collected using the manual for environmental DNA research from the eDNA Society⁴⁵. Water samples were collected at each site and filtered (typically 1 L per sample, with two replicates) to capture environmental DNA on membrane filters. DNA was then extracted from the filters using a standardized protocol. Fish DNA was amplified using MiFish primers targeting a fragment of the mitochondrial 12 S rRNA gene. The resulting amplicons were sequenced and used to infer species composition at each site. Data obtained from the same sampling site at different depths and times was analyzed as independent surveys. Genus-level data were used because 89.92% of OTUs (963/1,071 OTUs) were assigned to the genus level, whereas only 63.03% of OTUs (675/1,071 OTUs) were assigned to the species level, particularly for sequences with low read numbers. Undefined genus reads were still obtained; however, DNA sequences that could not be assigned to genus level were omitted from the analyses.

Rarefaction

The data in the ANEMONE database included two types of eDNA data due to the different methods used to obtain sequence reads (qualitative vs. quantitative). The eDNA data in 2017 and 2018 were obtained using universal polymerase chain reaction (PCR) primers (MiFish-U/E) for the metabarcoding of environmental DNA (eDNA) from fish⁴⁶. Because the concentrations of PCR inhibitors depend on certain uncontrollable conditions of the samples⁴⁷, this method cannot be used to estimate the amount of eDNA. This method is referred to as the qualitative method. eDNA data from 2019 onwards were obtained using an expanded method developed by Ushio et al., to detect fish biomass²³. In the expanded method, internal standard DNA is added to eDNA samples from the field, and metabarcoding analysis is performed; therefore, we can quantify the eDNA copy number of fish using the copy numbers from the sequence reads of the added standard eDNA as a calibration of the absolute abundance. This method is referred to as the quantitative method. The quantitative method is more beneficial, as it allows us to obtain more accurate information on fish abundance/biomass; however, a small amount of eDNA may be missed because standard eDNA interferes with the eDNA sampled in the field. Consequently, with regards to the data collected as presence/absence data, the quantitative method may have a limited detection capability compared to the qualitative method. It is possible that the species richness estimated using the qualitative method will be greater than that estimated using the quantitative method. To avoid misinterpretation due to the method difference, the sensitivity difference between the methods was corrected.

To solve this problem, a rarefaction method was used. The rarefaction method can be applied to downsize larger samples until they are the same size as the smallest sample⁴⁸. The rarefaction curve indicates the expected genus richness in the random subsamples of the sample from the community. The expected genus richness calculated using rarefaction should be corrected for differences in DNA detection sensitivity. Genus richness was estimated after rarefaction by subsampling sequencing reads to a standardized depth across samples, thereby controlling for differences in sequencing effort. Presence–absence of each genus was then determined based on the rarefied read data, and genus richness was calculated as the number of detected genera in each survey. To accomplish this, the ‘rarefy’ function of the vegan package (version 2.6.4) in R (version 4.3.2) was used. Note that no significant difference was observed between the qualitative and quantitative methods after rarefaction (Wilcoxon signed-rank test, W = 8732, p = 0.6852).

Data analysis

To assess the inter-regional fish LDGs, Japanese waters were divided into two regions: the Mainland area and the Ryukyu area (Fig. 1). The Ryukyu area includes the Ryukyu Islands (Fig. 1), where a different faunal structure of the Pacific coast along the Japanese Archipelago has been reported¹³, and specific local populations are formed^{49, 50, 51}. Therefore, the Ryukyu region was treated separately from the other regions, although we also analyzed the pooled data for all surveys (see Figure S1 and Table S2). The seasonal changes in LDGs in each region were also considered. A generalized linear model (GLM) was constructed to demonstrate the fish genus richness in each area, and the fish genus richness after rarefaction was settled on the response variables to correct the difference between qualitative and quantitative methods. In the models, latitude was chosen as an explanatory variable to explore LDGs, and the categorical variable ‘season’ was included with the four levels: winter (used as the reference category) from December to February, spring from March to May, summer from June to August, and autumn from September to November. The interaction between latitude and season was also included in the model. For the GLM analysis, a gamma error distribution with the identity-link function was used to account for the fact that the response variables were positive real numbers. We also tested a log link function and obtained similar results (Mainland area: AIC = 6806.3 for the identity link and 6810 for the log link; Ryukyu area: AIC = 1141.8 for the identity link and 1141.7 for the log link). Model assumptions were evaluated using diagnostic plots, including residual analyses. No clear residual pattern was observed in the Mainland area, whereas variance tended to be slightly larger at higher latitudes in the Ryukyu area (Figure S3). This may be due to bias resulting from the limited number of sampling sites. To determine whether the results were due to fish movement, we applied the same GLM method to inland waters where the sampling sites were separated.

Lastly, we examined whether the seasonal pattern differences in the LDGs were due to fish that are unique to the season (vagrant fish carried by ocean currents from outside regions) or to migratory fish. A season-specific genus was defined as a genus observed only in one season (e.g., the summer-specific genus at a sampling site was present in summer but absent in the other seasons). For season-specific fish, the ratio of survey sites with more than one season-specific genus to the total number of survey sites was investigated in Mainland area (i.e. excluding Ryukyu area) in each season.

In order to see the impact that migratory fish have on the LDGs patterns, the seasonal migration patterns of five genera of fish that are designated as useful fish in Japanese resource management by Japanese Fisheries Agency – Cololabis, Sardinops, Scomber, Seriola and Trachurus – were visualised in each season. These are genera widely distributed in Japan and have relatively well-known seasonal migration patterns (northwards in summer and southwards in winter)^{36, 52, 53, 54, 55, 56}.