Environmental DNA analysis as a non‐invasive quantitative tool for reproductive migration of a threatened endemic fish in rivers

Abstract Quantitative information regarding reproduction is essential for conserving endangered animals; however, some conventional quantitative methods can be damaging to the target population and their habitats. In the present study, the reproductive migration of a threatened endemic fish, three‐lips (Opsariichthys uncirostris uncirostris), was non‐invasively monitored by quantitative PCR of species‐specific environmental DNA (eDNA), the usefulness of which has been not sufficiently explored. Water sampling and from‐shore visual inspection were performed weekly along a tributary of Lake Biwa (Japan), where adult fish seasonally migrate upstream to reproduce as well as at lake sites near the river mouth. Species‐specific eDNA was collected at all locations at times when the fish were visually observed and at certain sites where the fish were not observed. Log‐transformed individual counts from visual inspection were positively correlated with log‐transformed eDNA concentration in the river sites, indicating that eDNA analysis can be a reliable quantitative tool for fish abundance in rivers. Furthermore, distance from the lake did not influence eDNA concentration, suggesting that eDNA transport by river flow had a negligible effect on eDNA quantification. Both eDNA concentration and individual counts gradually increased from May–July, and decreased in August. Importantly, eDNA analysis showed that the fish occupied more habitats in the peak reproductive season and stayed for longer time at any given site. An additional underwater survey confirmed unexpected eDNA detections as true positives. eDNA analysis has great potential to quantitatively monitor reproductive fish migrations under certain conditions.


| INTRODUC TI ON
Information regarding reproduction is essential for effectively and efficiently conserving rare and endangered animal species (Sutherland, 1998). However, since they often involve capturing animals and/or invading breeding grounds for direct observation during breeding seasons, conventional methods of surveying reproductive ecology are potentially damaging to the target species and its habitat. In addition, conventional survey methods require experts to be present for capture and/or species identification; thus, high financial costs can prevent the long-term monitoring of target species. Therefore, non-invasive, accessible approaches for studying reproductive ecology are required. Recently, environmental DNA (eDNA) analysis was proposed as a non-invasive, quick approach to monitoring aquatic macro-organisms (Ficetola, Miaud, Pompanon, & Taberlet, 2008;Jerde, Mahon, Chadderton, & Lodge, 2011;Lodge et al., 2012). This technique can reveal the distribution of target species by detecting species-specific DNA fragments in water samples. eDNA analysis can be performed non-invasively and simultaneously at several sites with more repetition than conventional methods, such as capture and direct observation, because the only fieldwork necessary is the collection of water samples (Darling & Mahon, 2011;Nakagawa et al., 2018;Yamamoto et al., 2016).
To assess whether animal density can be predicted by eDNA analysis, we selected a cyprinid fish, three-lips (Opsariichthys uncirostris uncirostris, Temminck et Schlegel), in a river as a model to test whether eDNA analysis can quantitatively monitor fish populations ( Figure 1). This fish grows in lakes and large adult individuals (14-25 cm standard length; 2 + -4 + age) only seasonally migrate upstream into tributaries for reproduction (Nakamura, 1951;Tanaka, 1970). Therefore, the body size of individuals in rivers is similar, which minimizes any possible effect that size may have on eDNA shedding rate (Maruyama et al., 2014). In addition, large three-lips individuals are easy to count from land (Imamura, 2014), enabling accurate abundance estimation by visual survey. In addition, a quantitative monitoring technique is required for the conservation of this unique piscivorous Japanese cyprinid subspecies, endemic to lakes Mikata and Biwa. The native populations are listed as vulnerable in the Red List specific to Japan (Japan Ministry of the Environment, 2010), because the Lake Mikata population is considered extinct, and the catch in Lake Biwa has been decreasing continuously since the 1980s (Tsunoda, Urano, & Ohira, 2015). In spite of this, the patterns of reproduction or migration of this species have not been studied since Tanaka (1970) investigated the age structure and growth rate of its spawning populations in the 1960s.
The objective of this study was to clarify the pattern of three-lips reproductive migration along a tributary of Lake Biwa by monitoring change in species abundance estimated by eDNA analysis as well as riverside and lakeside visual surveys. eDNA was validated as a method for quantitative fish monitoring in rivers. Environmental F I G U R E 1 Underwater photographs of a threatened endemic fish, three-lips (Opsariichthys uncirostris uncirostris), in a tributary of Lake Biwa, Japan factors affecting estimates of fish abundance by eDNA analysis were also explored.

| Study sites
We selected the Chinai River, a tributary of Lake Biwa in central Japan, as a model site for study ( Figure 2) because three-lips was reported to migrate along this river in greater numbers than along the other tributaries of Lake Biwa when the fish was still abundant (Nakamura, 1951). According to our 2016 preliminary survey, during its reproductive season, the fish is now most abundant in this river among the 12 largest tributary rivers on the west side of Lake Biwa, probably because unlike most other tributary rivers that dry up seasonally in the summer, this river maintains surface water all yearround. There is a fish trap spanning the width of the watercourse, and there are 11 small dams (0.22-1.17 m high) on the Chinai River, some of which might be barriers to the migration of fish.
Nine study sites along the Chinai River (180-5,800 m from the river mouth) and two sites on the lakeshore (within 100 m to the north and south of the river mouth) were selected ( Figure 2). All the study sites near the trap or dams (n = 6) were at least 3 m upstream from the trap or dams. The other river sites (n = 3) were 3 m upstream of the borders of different types of riverbed substrate (sand or gravel). The width and the depth of the river were 5-20 m and 15-50 cm, respectively, in the study area.
The field survey was conducted weekly from 10th May to 21st September 2017 (20 weeks) to cover the three-lips reproductive migration season (Nakamura, 1951). All the 11 sites were visited in the day time within one day once per week for water sampling, visual inspection and other measurements described below. Due to flooding, the field surveying was canceled four times. Thus, each site was visited 16 times in total, while some samples in the upper reaches of the river were not analyzed in off-peak seasons to reduce costs (see Section 3).

| Quantitative environmental DNA analysis
For eDNA analysis, 500 ml water was collected using a polypropylene bottle at each site and filtered on site using a glass fiber filter (Whatman GF/F, 0.7 µm pore size, GE Healthcare, Chicago, US) and a plastic filter funnel (ASONE, Osaka, Japan) within 90 min of sampling to ensure that eDNA decay was less than 10% (Barnes et al., 2014;Maruyama et al., 2014). To provide negative controls, distilled water was filtered at the beginning and end of each day. Immediately after each sample was filtered, 70% ethanol was added to the samples for eDNA preservation and the samples were preserved in portable freezers (Minamoto, Naka, Moji, & Maruyama, 2016). Before use, all the sampling and filtering equipment was washed with a 10% bleach solution to remove any residual DNA fragments and rinsed with distilled water to remove residual bleach. The sampling bottles were not reused over the course of each day. DNA was extracted from 185 samples (including 32 negative controls) using a DNeasy Blood & Tissue Kit (Qiagen, Hilden, Germany) in combination with a spin column (EP-11201, Gene Design, Ibaraki, Japan) following the method suggested by Miya et al. (2015) with slight modifications. The filter was tightly folded into a small cylindrical shape and placed into a spin column from which the attached silica gel membrane had been removed in advance. The spin columns were centrifuged at 6,000 g for 1 min to remove excess ethanol from the sample. 100 µl buffer AL and 10 µl proteinase K were mixed with 200 µl Milli-Q water to increase the volume enough to penetrate all the filter. The mixture was dispensed onto the filter in each spin F I G U R E 2 Locations of the survey sites (open circles) along the Chinai River (n = 9) and Lake Biwa shore (n = 2) in central Japan. Solid diamonds indicate positions of a river-crossing fish trap (n = 1, the closest from the lake) and small dams (n = 11). All the survey sites located near the trap or dams (n = 6) were settled 3 m upstream from the trap or dams Lake Biwa Zoomed area Chinai River Lake Biwa 0 1 km N column, and the spin columns were incubated for 30 min at 56°C.
After incubation, the spin columns were centrifuged at 6,000 g for 1 min to collect DNA and the eluted filtrate was transferred to a new 2.0 ml collection tube. To recover residual DNA on the filter, 200 µl TE (Tris-EDTA) buffer was added to each filter and the filter was incubated for 1 min at room temperature before being centrifuged at 6,000 g for 1 min. The upper part of the spin column containing the filter was removed from the 2.0 ml collection tube, and the first and The quantitative real-time PCR (qPCR) was performed in triplicate for each eDNA sample. To quantify the number of three-lips D-loop genes in each 2 μl eDNA template, we performed qPCR in triplicate simultaneously using a dilution series of standards containing 60, 600, 6,000, and 60,000 copies of the target sequences as standards in all qPCR assays. The standards were provided by a commercial service (Standard Genes, Eurofins Genomics K.K., Tokyo, Japan) in which the target sequence was cloned using pEX-K4J1 vector. The product was delivered as plasmids and used as the source of the dilution series. The solution of standard was stocked in a −20°C freezer at the concentration of 300,000 copies/µl, and the dilution series were prepared freshly for each PCR experiment.

| Visual inspection and measurements of environmental conditions
Immediately after water sampling at each site, the number of individual three-lips within a 30 m stretch upstream of the water sampling site was determined by visual inspection from the riverbank, bridge, or lakeshore. The same researcher (KS) performed the visual inspections throughout the study. The observer went into the water only when conclusive species identification was not possible from the shore. When three-lips were found in groups, they were observed carefully for 10 min to determine whether they were spawning.
Afterward, we measured the pH of the water using a compact pH meter B-712 (Horiba Ltd., Kyoto, Japan), conductivity using a compact meter B-771 (Horiba Ltd.), water temperature using a digital meter ND-X (Marukan Ltd., Osaka, Japan), current velocity using a propeller current meter CR-11 (Cosmo Riken Ltd., Osaka, Japan), and coordinates (latitude and longitude) using an iPhone 7 (Apple Inc., Cal., US) with a Google Maps application (Google Inc., Cal., USA).
The type of riverbed substrate over each 30 m observation site was classified by eye as gravel or sand during the first survey on 10th May 2017. No other types of substrate were observed within the study sites. In the two lake sites, lake bottom was sandy as far as the eye could see.
At the beginning of the apparent peak of the reproductive season, an additional underwater survey was conducted by drift snorkeling. The underwater survey was done on the next day of the 10th survey (i.e., 13th July 2017) to avoid possible cross-contamination through the snorkeling equipment and disturbance to the fish. An experienced observer (AM) swam down the river from the uppermost site (5,800 m from the lake) to the fifth-highest site (2,350 m from the lake), where three-lips had not been observed during visual inspections. The observer was able to identify any individual fish larger than ~10 cm standard length in the whole area between the both shores. Underwater survey was not appropriate for counting the three-lips, which tended to flee from observers in the water; the underwater survey was conducted only to investigate possible false positive eDNA analysis results in the upper reaches of the river where the fish were not observed from the riverbank or bridge.

| Data analyses
Since measuring the distribution and abundance by quantitative eDNA analysis and visual inspection usually gives zero-inflated data The seasonal changes in distribution and abundance of threelips were examined in two steps. First, using detection data (positive/negative) from the river, the relationship between the number of dates on which fish were detected and distance from the lake at each river site was examined separately for the two methods using Spearman's rank correlation test. Second, using quantitative data from the river, changes in estimates of the fish quantity (eDNA concentration and visual counts) were examined, before and after the first observation of spawning behaviors and separately for the two methods, by LM with time (weeks after the first survey), distance from the lake (m) and an interaction term for these two continuous variables as explanatory variables. Estimates of three-lips abundance in the lake were examined by LM in the same manner, but with time as the only explanatory variable.

| Validity of eDNA analysis in comparison to visual inspection
There was no amplification of the three-lips species-specific eDNA in any distilled water samples filtrated before and after test water samples or in any negative control using qPCR. Thus, inter-sample contamination was considered to be negligible throughout the analysis. DNA sequencing after qPCR confirmed that the sequences of all 93 amplicons examined were from three-lips.
eDNA analysis detected three-lips eDNA in 89 out of 153 samples, whereas the fish were only visually observed 32 times during 153 inspections (Table 1) When the 89 positive eDNA data points were analyzed, it emerged that the probability of observing three-lips during visual inspection was significantly correlated with the log 10 -transformed eDNA concentration (GLM with a logit link function, z = 4.12, p < 0.001; Figure 3) and was significantly different between the river and lake sites (z = 3.18, p < 0.01).
When the 32 data points positive in both analyses were analyzed, the log 10 -transformed individual number of three-lips observed by visual inspection was positively correlated with the log 10 -transformed eDNA concentration at the river sites (LM, R 2 = 0.574, p < 0.001; TA B L E 1 Relationship between the detection of three-lips by the environmental DNA (eDNA) analysis and visual inspection, which were concurrently conducted 16 times during May-September 2017 at nine river sites and two lake sites eDNA analysis

Visual inspection
Positive 32 0 Negative 57 64 Note. Numbers of data sets, each of which is derived from an eDNA sample and an inspection at a site on a day, are shown. Figure 4), but not at the lake sites (R 2 < 0.1, p > 0.5). In the river, the individual counts from visual inspection (N) can be predicted from eDNA concentration (C) as log 10 N =-0.842 + 0.6131 × log 10 C.
Water temperature and riverbed substrate were selected as explanatory variables by stepwise variable selection using AIC. The residuals of the regression (Figure 4) were significantly explained by the riverbed substrate (LM, t = 3.16, p < 0.01; Figure 5), but not by water temperature (t = 1.37, p = 0.182). The residual was larger at the sites with sand substrate than at the sites with a gravel substrate. In other words, more three-lips individuals were counted than expected based on the eDNA concentration at the sites with sand substrate than at the sites with gravel substrate.

| Seasonal changes in distribution and abundance monitored by eDNA analysis and visual inspection
The detection data from the eDNA analysis and visual inspection along the river indicated that the target fish expanded its distribution upstream from the river mouth for 1-2 months ( Figure 6).
The number of dates on which fish were detected decreased with site distance from the lake, both for eDNA analysis (ρ = -0.962, p < 0.001) and visual inspection (ρ = -0.957, p < 0.001). eDNA analysis and visual inspection recorded fish up to 3,140 and 2,350 m from the lake, respectively. Interestingly, in all cases, the fish presence at each site was detected by eDNA analysis before visual inspection across the entire survey period. During the additional underwater survey from the uppermost site (5,800 m from the lake), fish were first detected at 3,160 m from the lake. At the two study sites in the lake, the target fish was detected by both methods; however, it was observed four times only over 32 visits by visual inspection.
Spawning behaviors were first observed by eye on 19th July in the lower reaches of the river and continued for three weeks ( Figure 6). The quantitative data from the two methods in the river showed significant increases in abundance with time, until the beginning of spawning behaviors ( Table 2). The slope of the increase in eDNA concentration significantly decreased with distance from the lake, as shown by the interaction term (distance × time), and hence, the peak tended to be higher in the lower reaches of the river than in the upper reaches. After the first observation of spawning behaviors, both the eDNA concentration and individual counts significantly decreased with time. The slope of this decrease was steeper at the sites closer to the lake.
In contrast, the results of the two methods did not match in the lake. The eDNA concentration significantly increased with time until the beginning of spawning behaviors, but individual counts did not ( Figure 6; Table 2). After the first observation of spawning behaviors, neither the eDNA concentration nor the individual counts changed with time. Consequently, eDNA was collected until the end of the survey period in the lake sites only.

| eDNA concentration as a quantitative indicator of fish abundance in rivers
eDNA analysis detected three-lips eDNA at all the locations and times when the three-lips was visually observed, without any false The results of visual inspection (P: positive, N: negative) of three-lips (left Y-axis, plots) as a function of its species-specific environmental DNA (eDNA) concentration (log 10transformed). Open circles and crosses indicate river and lake survey sites, respectively. These plots are vertically staggered to minimize overlapping. 89 positive eDNA data points were used ( Table 1). The solid and dotted lines (right Y-axis) indicate probabilities of detection by visual inspection estimated by logistic models for river and lake survey sites, respectively Concentration of eDNA (log, copies/L) F I G U R E 4 Scattered plot (a) and log 10 -log 10 plot (b) of individual counts per 30 m of three-lips by visual inspection and its speciesspecific environmental DNA (eDNA) concentration. Open circles and crosses indicate river and lake survey sites, respectively. 32 positive visual inspection data points were used (Table 1). The solid and dotted lines indicate linear models for river and lake sites, respectively

Individual counts by visual inspection
Individual counts by visual inspection (log) negative results. This result strongly supports the detection power of eDNA analysis with species-specific primers in the field that many previous studies have suggested (Jerde et al., 2011;Lodge et al., 2012;. Target eDNA was also detected in many samples taken when and where the fish was not visually observed. This does not necessarily imply that eDNA analysis is a more powerful detection tool than visual inspection; false positive results can also explain this situation (Yamamoto et al., 2016). Thus, we checked the sequence of qPCR amplicons and possible inter-sample contamination. In addition, we performed an F I G U R E 5 The residuals of the linear model (on visual counts by eDNA concentration) for river survey sites (solid line in Figure 4b) as functions of environmental factors: distance from the lake, pH, conductivity, water temperature, log 10 -transformed current velocity, and riverbed substrate. Horizontal solid lines indicate residual = 0. In the box plot, bold lines, boxes, and bars indicate medians, quartile ranges, and maximum-minimum ranges, respectively Residuals of the regression F I G U R E 6 The weekly data of threelips species-specific environmental DNA (eDNA) concentration (log 10 -transformed; left Y-axis, bars) and individual counts per 30 m by visual inspection (log 10transformed; right Y-axis, plots and lines) at nine river sites (light gray bars, 180-5,800 m upstream from the lake) and two lake sites (dark gray bars). Open circles on the plots indicate observation of spawning behaviors. Spaces without plots indicate skipped surveys due to heavy rain (3 days) or to save cost (others). "+" and "-" indicate positive and negative detection of the fish by underwater survey, respectively In addition, the regression analysis showed a higher probability of visual detection in the river than in the lake by one order in eDNA concentration. Since the depth gradient makes visual detection difficult in the lake, and unpredictable water current and eDNA diffusion may have made eDNA analysis inaccurate in the lake (Dunker et al., 2016;Eichmiller, Bajer, & Sorensen, 2014;Ghosal, Eichmiller, Witthuhn, & Sorensen, 2018), this difference is not surprising.
Researchers should pay attention to the unreliability of conventional data as well as to that of eDNA data (Chambert et al., 2018). It is also notable that both positive and negative visual detection results were obtained in the range of eDNA concentration widely from 10 2 to 10 3.7 copies/L (see Figure 3), probably because of an uneven detection efficiency of both methods, depending on the environmental heterogeneity within the river.
The scatter plot of individual counts by visual inspection and eDNA concentration (Figure 4a) suggests that the estimation of fish abundance is difficult or impossible, even under optimal conditions such as in our study, where large adult individuals (14-25 cm in standard length; 2 + -4 + age) of the target fish only seasonally migrate upstream into tributaries for reproduction (Nakamura, 1951;Tanaka, 1970). However, the log-log plot and the associated statistical analyses strongly support the possibility of accurate abundance estimation in the river, which implies that eDNA quantification can detect the order of magnitude variations in fish abundance under desirable conditions. Thus, quantitative eDNA analysis can be a good indicator of animal density, provided order-level resolution of quantification is understood, as the present study shows. Importantly, the residual analyses also indicated that most environmental conditions examined do not affect abundance estimations at the log scale, although some of these factors (i.e., pH, conductivity, and temperature) may have potential to affect abundance estimation when their ranges are wider than in our study. Notably, the distance from the lake did not affect the estimation. This result suggests that eDNA transport from the upper reaches of the river has a negligible effect on quantitative analysis at the spatial scale of this study; significant eDNA transport would have yielded a negative relationship between distance and eDNA concentration as long as eDNA concentration is considerably high in the upper reaches (Jane et al., 2015;Rice, Larson, & Taylor, 2018). However, a water current of 10-10 2 cm/s ( Figure 5) carries eDNA roughly 360-3,600 m downstream per hour, during which time only 5%-15% eDNA decays under experimental conditions (Klymus et al., 2015;Maruyama et al., 2014). These facts suggest that some mechanisms other than DNA decay, such as sedimentation and substrate trapping, may decrease the amount of eDNA detectable in the field (Shogren et al., 2016(Shogren et al., , 2017. Further studies are required to understand the quantitative effect and mechanisms of eDNA flow in lotic water (Seymour et al., 2018); however, our findings would support the use of quantitative eDNA analysis in lotic environments. In contrast, the use of eDNA for the quantification in the lake was not validated in this study, because the depth gradient makes counting fish by eye difficult and also because unpredictable water currents and eDNA diffusion may have made the eDNA analysis inaccurate, as stated above (Dunker et al., 2016;Eichmiller et al., 2014;Ghosal et al., 2018).

| Seasonal distribution and abundance of threelips
Seasonal migration of three-lips into rivers had not been studied since Tanaka (1970) reported the age structure and growth rate of its spawning populations. Thus, we had no a priori knowledge of seasonality in upstream migration or distribution of three-lips in its peak breeding season. This study used eDNA analysis and visual inspection to show that the target fish expands its distribution upstream from the lake gradually, and remained in the river for 3-4 months.
An important difference between the results of the two methods was eDNA analysis showed a larger distribution at the peak of the reproductive season as well as longer residence time at any given TA B L E 2 Coefficients (±standard errors) of linear models (LM) to explain the changes in three-lips species-specific environmental DNA (eDNA) concentration and individual counts of the fish by visual inspection (visual)
site. Such novel knowledge provided by eDNA analysis may have important implications for the conservation of rare and endangered animal species, for example, when authorizing appropriate fishing areas and seasons.
The quantitative data from both methods in the river showed a gradual increase in abundance followed by a decrease at each site, which indicated the peak season of the three-lips reproductive migration clearly. In addition, it was shown that more individuals use the lower reaches of the river than the upper reaches. The eDNA concentration in the lake suggests that the fish gather at the river mouth from May until July. Afterward, unlike at river sites, the eDNA concentration remained high, which suggests the fish remains near the river mouth after reproduction. This is also novel information provided by eDNA analysis that may have conservational implications. We propose that several other fish species with similar life histories living in Lake Biwa and its tributaries could be quantitatively monitored using eDNA analysis. This would result in further novel findings obtained via non-invasive methods.

| CON CLUS ION
In the present study, we monitored the reproductive migration of a threatened endemic fish, three-lips, by quantitative eDNA analysis accompanied by visual inspection. Species detection by eDNA analysis was shown to be accurate and demonstrated a larger distribution and longer residence time of the target fish in the river. In addition, we showed that eDNA analysis, usefulness of which has been considered limited in this context previously, can be a good quantitative tool for fish in rivers at the log scale. Estimating fish abundance using eDNA concentration demonstrated a gradual increase and decrease in the number of fish present. Thus, we propose that, under certain conditions, eDNA analysis has greater potential than conventional methods for quantitatively monitoring the reproductive migration of fishes without damaging the target fish population or habitats.

ACK N OWLED G M ENTS
We are grateful to S. Okui, S. Fujiwara, J. Takemura and H. Sawada for their support in the field, S. Okayama for underwater photographs, and S. Tada for instructing the DNA analysis. Our thanks go to S. Tsuji, K. Shigeta. M. Kawakami and other students of Ryukoku University for discussion. This study was supported by JSPS KAKENHI (No. 16K00630, 15K14792).

CO N FLI C T O F I NTE R E S T
There are no conflict of interests to declare.

AUTH O R CO NTR I B UTI O N S
A.M., H.Y. and A.I. conceived the ideas; A.M. and K.S. designed the field survey. K.S. collected and analyzed the data under supervision by A.M., K.W. and H.Y. A.M. led the writing of the manuscript. All the authors contributed critically to the drafts and approved the submission.

DATA ACCE SS I B I LIT Y
Data available from the Dryad Digital Repository: https://doi. org/10.5061/dryad.5hb0fb8.