Unprecedented rates of biodiversity loss and ecosystem function necessitate the use of rapid, efficacious, and cost-effective biomonitoring tools. The combination of autonomous samplers and high throughput sequencing (i.e., “metabarcoding”) of environmental DNA (eDNA) samples enables characterization of entire communities at high frequency and can be an important tool for conservation and management, allowing researchers to track fluctuations in biodiversity. We deployed two autonomous samplers at two U.S. Geological Survey streamgage sites in the upper Snake River (Wyoming and Idaho, USA) to collect eDNA samples from July-September 2021 and 2022 to characterize fish diversity. We used a probabilistic approach to evaluate the effects of water temperature, water discharge, filter pore size, water volume filtered, number of samples collected, timing, and primers on the probability of detecting eDNA from fish species known to be present. We detected eDNA from 13/15 species present in these areas of the Snake River. Overall, we did not find evidence that filter pore size, water volume filtered, water discharge, and water temperature affected the probability of detecting fish species’ eDNA. By contrast, primers and sampling day affected fish detection probabilities, indicating that primer choice and sampling day can either over- or under- estimate species diversity. These results indicate that users would ideally consider sampling on non-consecutive days and which primer set will maximize species detections.

Autonomous samplers, biomonitoring, eDNA, metabarcoding, USGS streamgage

Biodiversity monitoring informs species conservation, ecosystem service management, and progress toward major national and international policy initiatives (e.g., Díaz et al. 2019; Aichi Biodiversity Targets 2020). However, there is a mismatch between the scale of biomonitoring needed in our rapidly changing world and the biomonitoring tools currently available. Conventional biomonitoring tools (e.g., nets and traps) are limited to short temporal windows, low sampling frequencies, and can be restricted to a subset of species present in the environment (Moussy et al. 2022). These shortcomings can result in a misrepresentation of biodiversity and inefficient allocation of limited resources to achieve conservation outcomes. There is a need for the development and implementation of additional tools that overcome these limitations.

Environmental DNA (eDNA) metabarcoding analysis is a biomonitoring approach that allows for biological communities to be described from short sections of DNA that occur in easy-to-collect environmental samples. The approach is accurate, rapid, cost-effective, non-invasive and can be consistently and broadly applied (Ruppert et al. 2019). Nevertheless, eDNA metabarcoding, like any other tool, provides imperfect representation of biological life present due to constraints imposed by a myriad of field and laboratory factors. Here, we focus on the implementation of new technology and sample design considerations to help overcome one of these factors that has been generally neglected in previous studies—temporal heterogeneity of eDNA in the environment (e.g., Mathieu et al. 2020; Hallam et al. 2023).

The amount of eDNA available to be sampled at a site is temporally variable because it is a product of dynamic biological and environmental interactions. Organisms release complex mixtures of DNA in the environment at inconsistent rates and amounts that are associated with taxonomy, size, activity, and phenology among other things (Barnes and Turner 2016). Abiotic factors and microbial activity break down this eDNA into smaller fragments while it is being adsorbed and transported in multiple dimensions (Jo et al. 2020). Therefore, an eDNA sample is often a snapshot of the subset of eDNA present at a specific time and space and is likely to provide incomplete representation of biological communities (Jensen et al. 2022). Multiple samples over time can be required for a more complete biomonitoring picture and to have confidence that eDNA-inferred changes in community composition across time points are indicative of species turnover rather than sampling bias. However, human resource limitations pose challenges to the increased temporal sampling needed to complete this picture.

Autonomous eDNA sampling instrumentation presents an opportunity to overcome human resource limitations for ecosystem monitoring over time and space. Autonomous eDNA instruments, like Monterey Bay Aquarium Research Institute’s environmental sample processor (ESP) platforms (Moss Landing, CA, USA), can collect and preserve dozens of eDNA water samples without a human operator (e.g., Scholin et al. 2017; Sepulveda et al. 2021; Preston et al. 2023). Thus, eDNA data can be acquired at sites and times that are less amenable to manual sampling, such as remote sites, nights, and hazardous environmental conditions. This technology has great promise to reduce biomonitoring misrepresentation associated with the temporal heterogeneity of eDNA in the environment because samples can be collected at finer temporal frequencies (e.g., sub-daily; Searcy et al. 2022) for longer durations (e.g., weeks to months) than most manual eDNA sampling efforts. However, continuous improvement of study design is needed to ensure temporal sampling frequency is appropriate for the system because sample number is still limited, and autonomous sampler deployment and analysis are costly.

Understanding about appropriate eDNA temporal sampling frequencies for biomonitoring is minimal relative to many eDNA workflow methods (Mathieu et al. 2020), likely because eDNA autonomous sampling is a newer tool and meaningful insight on this topic is largely limited to field studies rather than laboratory experiments. The number of temporal studies is increasing, but is still largely limited to longer (i.e., months to seasons) rather than shorter time scales (i.e., hours to weeks) (Bista et al. 2017; Beentjes et al. 2019; Handley et al. 2019; Hayami et al. 2020; Mariani et al. 2021; Sales et al. 2021; Hervé et al. 2022). Sampling at longer time scales can capture eDNA from taxa that are seasonally present (e.g., migration) and that have increased activity associated with phenology (e.g., reproduction). By contrast, sampling at shorter time scales is more likely to capture eDNA from low density species that quickly degrades and whose activity and habitat use vary diurnally. For example, Jensen et al. (2022) showed that fish species eDNA detections were highly dynamic throughout a 32-hour period, with eDNA-inferred species richness varying substantially at an hourly rate with some species having a greater number of detections mid-day while other species had a greater number of detections at night. Thus, sampling at shorter time scales that more accurately represents community composition at a site may be critical to increasing confidence in species turnover inferred from eDNA samples collected at longer time scales, which is a primary objective of many biomonitoring programs (Bálint et al. 2018). As resources limit the amount of sampling that can be done for biomonitoring, trade-offs between short- and long-term sampling periods should be considered.

To gain insight on temporal sampling tradeoffs for eDNA biomonitoring programs, we leveraged existing eDNA samples that were collected for invasive species surveillance by autonomous samplers at sub-daily frequencies for three months from summer (July) to autumn (September) seasons in 2021 and 2022. The autonomous samplers were deployed at U.S. Geological Survey (USGS) streamgages immediately downstream of two reservoirs in the Snake River (Idaho and Wyoming, USA). These waters contain fish species known to exhibit variation in daily and seasonal activities, such as cutthroat trout (Oncorhynchus clarkii) that reproduce in the summer and lake trout (Salvelinus namaycush) that reproduce in the fall (Gunn 1995; Henderson et al. 2000). We used eDNA metabarcoding approaches on these existing samples to describe fish community composition in the upstream reservoirs. We predicted that 1) fish eDNA detections will vary sub-daily, daily, and seasonally and 2) an increased number of samples collected per day will increase the number of fish species eDNA detections. We also evaluated how temporal sampling tradeoffs could be influenced by other study design decisions like filter pore size and primer choice while accounting for covariates like water temperature and water volume filtered. Our analyses highlight the importance of sampling design choices in the field and lab when using eDNA samples for biodiversity monitoring.

We sequenced a total of 176 field samples with the number of samples collected per day ranging from one to six, eight mock community samples, and 20 negative controls (including field, extraction, PCR, and sequencing blanks). In 2021, we sequenced 45 samples from Jackson Lake and 45 samples from Palisades Reservoir collected over 12 days. In 2022, we sequenced 46 samples from Jackson Lake (n_MiFish = 46, n_ACMDB = 46) and 40 samples from Palisades Reservoir (n_MiFish = 40, n_ACMDB = 40) collected over 16 days (breakdown of samples collected at each site per day listed in Suppl. material 1: table S3). Across all datasets, five classes of eukaryotes were detected (Actinopteri, Aves, Mammalia, Bacillariophyceae, and Insecta) with a total of 8,058,780 reads. We retained only fish (Actinopteri) detections, resulting in 63 unique species for the 2021 dataset (2,041,818 reads), 71 unique species for the 2022 MiFish dataset (3,837,411 reads), and 69 unique species for the 2022 AcMDB dataset (2,400,946 reads). Finally, we filtered the dataset further to include only species known to be in the Snake River, resulting in a total of 3,709,008 reads across 11 detected species (Table 1) for both sites and years in our final dataset.

Differences among filter pore sizes and optimized sampling

We found overlapping probabilities of detecting eDNA from the same number of fish species for each of the filter pore sizes, indicating no effect of filter pore size on results (Fig. 2). The number of samples needed to detect eDNA of most fish species present reached an asymptote beginning at 2–3 samples per day (Fig. 2). These results are consistent between sites and months.

Figure 2.

Detected species accumulation curves showing the point and interval estimates for the expected number of species detected from eDNA based on the number of samples collected for each site during each month in 2021. Colors indicate the three different filter pore sizes tested. Bars represent 95% confidence intervals.

Differences between primer sets

The two different primers were able to detect eDNA from a total of 13 out of 15 fish species known to occur (Table 1). Neither primer set was able to detect Catostomus platyrhynchus or Cottus beldingii. For one (C. platyrhynchus) of the two undetected fish species, which had no 12S sequence data available, as well as C. ardens, we were able to find specimens (i.e., fin clips from fish in the upper Snake River) to sequence the entire mitochondria. These mitochondrial sequences can be found under the GenBank submission number BankIt2662303 (#s PP116420–PP116425).

The MiFish primer set was able to detect eDNA from three species (C. bairdii, G. atraria, and Salvelinus namaycush) that AcMDB was unable to detect. Interestingly, the AcMDB primers were able to detect C. bairdii and S. namaycush in the mock communities. By contrast, the AcMDB primer set was able to detect one additional species (C. ardens) in the field samples that the MiFish primer set did not. We found primer choice resulted in species-specific differences in detection using samples from 2022 to assess the difference in fish detection probabilities. Five fish species were detected with both primer sets; six species were detected with only one primer set (Fig. 3, Suppl. material 1: table S5). The AcMDB primer set had a greater probability of detecting eDNA from C. ardens and R. osculus, whereas the MiFish primer set had a greater probability of detecting eDNA from C. bairdii, G. atraria, O. nerka, R. cataractae, and S. namaycush (Fig. 3). AcMDB primers were able to detect O. nerka in a single sample and only at low levels (i.e., 3 reads in that sample).

Figure 3.

Detection probabilities of each primer for each fish species detected from eDNA in 2022. AcMDB primer detection probabilities are shown in purple for each species, and the MiFish primer detection probabilities are shown in pink for each species. Bars represent 95% confidence intervals.

Effects of timing on sampling

We found that detection probabilities varied by day but lacked a consistent or obvious temporal pattern. Days with 95% credible intervals that do not overlap with zero indicate that the eDNA detection probabilities of all species were higher or lower on those days compared to the average day, implying that either more or fewer number of species’ eDNA will be detected on those days (Fig. 4). There was a single day (day 208; July 27, 2022) in July at the Jackson Lake streamgage where fish species’ eDNA detections were lower than average. Similarly, there was a single day (day 208) in July at the Palisades Reservoir streamgage with below average fish species detection probabilities. This incidence of below average detection probability at Palisades Reservoir was immediately followed by a day of above average detection probability (day 209; July 28, 2022). There were also two consecutive days in September at the Palisades Reservoir streamgages site when detection probabilities were higher than average (days 249–250; September 6–7, 2022). We did not find a difference in detection probabilities for any fish species between summer and autumn sampling.

Figure 4.

Plots of the detection probability random effect point and interval estimates for each sampling day by site for 2022 samples. Pink points highlight daily detection probability estimate intervals that do not overlap with zero. Bars represent 95% confidence intervals.

In light of global declines in biodiversity, it is important to understand how to integrate eDNA sampling into robust monitoring frameworks. Biomonitoring programs are challenged to balance tradeoffs in study design choices, such as sampling frequency and duration since human and funding resources are limited. Combining eDNA metabarcoding with autonomous samplers into a biomonitoring program provides a new way to reduce these limitations, but study design tradeoffs must still be considered and optimized to minimize the misrepresentation of communities. Here, we considered how sample timing frequency, filter pore size, and primer choice influence inferences of fish community composition. Taken together, results from this study show that, of the factors we considered, primer choice and sampling date had the greatest impact on fish species eDNA detections.

High frequency samples collected autonomously across months from USGS streamgage sites were able to detect eDNA from 13 of 15 fish species known to occur at two upstream reservoirs. Though these data provide an imperfect representation of the known fish assemblage, our results underscore the relatively minimal field effort needed for eDNA-based biomonitoring when using autonomous samplers. Traditional assessments can require several biological technicians employing multiple, potentially destructive methods (e.g., gill nets and electrofishing) at multiple sites and depths and can still result in biased findings. Comparable manual eDNA sampling would have required dozens of site visits. Our assessment required approximately six hours of field time at each of two sites for ESP deployment and retrieval. Technological advancements and standardization are likely to accelerate workflows such that eDNA autonomous platform-based programs can respond to the scale of biomonitoring needed in our rapidly changing world. An outstanding example of this promise is the use of uncrewed surface vessels fitted with autonomous platforms that were able to collect eDNA samples over a 4200-km, 29-day transit in the northeastern Pacific Ocean (Preston et al. 2023).

Technological advancements like autonomous platforms cannot be used to their full potential in biomonitoring programs if genetic reference libraries are insufficient. Inadequate genetic reference library coverage likely limited our ability to provide a complete picture of Jackson Lake and Palisades Reservoir fish assemblages. Of the two species’ eDNA we did not detect, one of these fish species had no 12S sequences available, whereas the other one had only six 12S sequences available. The use of multiple primers did not resolve this issue. The problem of incomplete reference libraries is nearly ubiquitous in eDNA biomonitoring efforts (Schenekar et al. 2020; Keck et al. 2023). Solutions include development of curated, regional databases tied to vouchered specimens (e.g., Dziedzic et al. 2023), expansion of and improvements to public reference databases (e.g., Westfall et al. 2023), or embracing taxonomy-free approaches (e.g., Pawlowski et al. 2021).

Our results emphasize that primer choice is one of the most influential considerations in eDNA metabarcoding study design, as has been demonstrated many times over (e.g., Hajibabaei et al. 2019). Even though both of our primers targeted the 12S region, the MiFish primers were able to detect eDNA from four more species than the AcMDB primers. Given that all but one fish species (R. osculus) detected with the MiFish primers had a detection probability greater than 0.6, most species’ eDNA should be detected with only a few samples collected per day given that the eDNA is available to be sampled. However, the ability of the AcMDB primers to detect an additional two species in the mock communities compared to field samples indicates that these primers may be more efficient at detecting fish DNA under ideal conditions (e.g., no inhibition and ample template). As eDNA can degrade quickly in the environment, shorter amplicons (i.e., < 200 bp) may be more easily recovered in eDNA samples (Hajibabaei et al. 2006; Meusnier et al. 2008). The MiFish primers target a ~170 bp fragment of the 12S mitochondrial region, whereas the AcMDB primers target a fragment nearly twice the length (~281 bp), which may have hampered our eDNA detection of some species in field samples. There was a benefit to using multiple primers as we were able to detect eDNA from one additional species, but the costs of additional analyses should be considered. Given the impact of primer choice on metabarcoding results, it is important to consider which primer set will optimize species detections to most accurately represent the community of interest.

Implementation of autonomous platforms facilitates high frequency eDNA sampling, which can provide insight into changes in an organism’s ecology. Indeed, previous research has shown that sub-daily measurements of eDNA concentrations may reflect changes in daily behavior and activity throughout a photoperiod (e.g., Searcy et al. 2022), and similarly, longer-term high frequency sampling may give insight into seasonal behaviors such as migration or reproduction (e.g., Tillotson et al. 2018; Hallam et al. 2023). Even when accounting for primers, the number of reads for each species was highly variable throughout the day (see example in Suppl. material 1: fig. S3), but because we did not have sample replicates for time of day, we were unable to test the effect of time of day on species detection probabilities. While other eDNA studies have shown that sub-daily sampling mimics diel behavior of fish species (e.g., Jensen et al. 2022; Searcy et al. 2022), our results did not show consistent patterns in the number of reads for any fish species throughout a given day (Suppl. material 1: fig. S3) and in fact, more closely resemble the sub-daily patchiness in detections seen in Searcy et al. (2022). Therefore, our results did not indicate diel differences in detection, even though several of these fish species are known to have diel variation in activity (e.g., Nowak and Quinn 2002). For example, some species of suckers and dace tend to exhibit more diurnal behavior, whereas some chub species have been shown to be crepuscular (Reebs et al. 1995). Similarly, rainbow trout are likely more diurnal in contrast to cutthroat trout (Nowak and Quinn 2002). Generally, a fish species’ diel behavior may vary depending on factors like temperature, age, and food availability. Lack of diel variation could be true or could be a function of sampling downstream of bottom-releasing reservoir dams, where hydraulics may mix eDNA and decouple ecological relationships between eDNA signal and timing at short temporal scales. Hallam et al. (2023) also found low dissimilarity in inferred fish community composition at sub-daily scales and suggested that eDNA mixing caused by tidal fluctuations may have muted diel patterns.

Sub-daily samples showed that adequate coverage of fish species was reached relatively quickly. For both sites and seasons, the sampling effort needed to optimize the fish species detections via eDNA reached an asymptote after three samples. Collecting more than three samples per day did not increase the probability of detecting more fish species’ eDNA, and therefore may not be worth the additional collection and analysis costs. Autonomous platforms have a limited number of samples that can be collected and stored before needing replacement. For example, the Dartmouth Ocean Technologies eDNA Sampler can collect nine samples and the Woods Hole Oceanographic Large Volume eDNA sampler can collect 12 samples (Table 1 in Hendricks et al. 2023). At our study sites, three samples per day were sufficient to describe detectable fish species while minimizing cost and human time and effort.

We found that spreading out eDNA samples across days rather than within days is more critical to minimizing misrepresentation of communities. While the probability of detecting eDNA from all fish species was consistent on most days, there were five days that clearly had lower or higher detection probabilities, and thus observed species diversity, compared to the average day. In fact, at the Palisades Reservoir streamgage, there were two consecutive days when eDNA samples showed large swings in detection probability and diversity relative to the average. Day 208 resulted in below average diversity followed by day 209 when replicates averaged higher than expected detection probabilities. Similar to Hallam et al. (2023), who used shorter- and longer-term temporal sampling schemes to describe fish communities detected by eDNA, none of the environmental variables considered here explained this variation. Therefore, until more informative covariates are identified, we are unable to identify the mechanism behind these fluctuations and cannot predict which eDNA sampling days will accurately represent the fish communities. These sampling day effects indicate that eDNA samples should not be collected on a single day, or even on consecutive days, but rather be temporally dispersed. Spreading out eDNA samples across days will reduce the likelihood of misrepresenting fish diversity and may even improve detection of new biological threats (Sepulveda et al. 2021). Simulations of targeted eDNA detections showed similar results: dispersing eDNA sampling across days may be more effective at detecting new invasions whose DNA is likely to be at low abundances compared to increasing the number of samples collected within a day (Sepulveda et al. 2021). Likewise, Hallam et al. (2023) provided similar guidance where they suggested eDNA biomonitoring should be spread across seasons rather than at a discrete time of year.

In contrast to other research assessing eDNA detections at a longer time scale (e.g., Hallam et al. 2023 and Searcy et al. 2022), we did not detect a difference of seasonal effect at either site, even when we account for the different primers used. In combination with the lack of statistical difference in eDNA detection probabilities for each species between seasons, these results indicate that the probability of detecting these fish species at the USGS streamgage site did not change between July and September even though their activity, relative abundance, and distribution in the reservoirs likely did. This may be a function of hydrodynamics and site features (i.e., reservoirs and dams). Water temperature and discharge were consistent during July and September relative to the rest of the year and did not appear to influence our results.

When using eDNA sampling to detect species of interest, it is important to consider which specific supplies (e.g., filter type and pore size) are needed to optimize the likelihood of detection. We found that the effect of pore size was minimal relative to other factors such as primer choice and sampling frequency. Previous research has shown mixed results where filter pore size for different filter types (i.e., MCE, Cellulose Nitrate, and Glass Fiber) may (Deiner et al. 2018) or may not (Bessey et al. 2020) have an impact on fish eDNA metabarcoding results. Similar to Bessey et al. (2020), we found no evidence of a difference in the probability of detection of each species due to filter pore size; however, we would likely only be able to detect large effects because there were only six paired sets of samples using each filter sizes of 0.45, 1.2, and 5.00 µm. These results held when we accounted for the differences in water volume filtered. Still, it is important to note that eDNA sampling is not a “one size fits all” approach and these results may not be applicable to all systems (e.g., Kumar et al. 2022). Before designing an eDNA biomonitoring program, site characteristics (e.g., turbidity) should be considered to optimize methods and supplies.

Metabarcoding of eDNA samples is becoming more commonly used and applied toward species detections and may be useful for future management (Hallam et al. 2023). Results from empirical studies like ours may help inform best practices for biomonitoring at USGS streamgages. In this study, we demonstrated the effectiveness of leveraging existing USGS streamgages for characterizing and monitoring fish diversity downstream of reservoirs. Reservoirs provide important services such as hydropower, flood control, and recreation, but can also be hotspots of invasive species introductions (Havel et al. 2005; Johnson et al. 2008). With more than 8,200 USGS streamgages located throughout the United States (Eberts et al. 2019), the incorporation of eDNA metabarcoding to characterize aquatic fauna could be a powerful way to monitor biodiversity. Results from this study may be useful in informing future sampling design for biomonitoring at USGS streamgages.

Finally, we demonstrate the usefulness of autonomous eDNA samplers to describe fish communities at USGS streamgages downstream of reservoirs. We showed that, of the factors we considered, primer choice and sampling day had the greatest impact on species eDNA detection probabilities. Autonomous platforms, like the ESP, are particularly useful in that they can collect many eDNA samples. One current limitation of many autonomous platforms is their inability to collect multiple replicates at the same time. Additional replicates collected in parallel, either autonomously or manually, could account for the impact of time of day on species detections. When autonomous platforms are used in biodiversity monitoring programs and budgets are limited, spreading out samples over the course of several days rather than taking several samples in a short period of time may be optimal.

We would like to thank Dr. Donald Zaroban (College of Idaho), Jesse McCane (Idaho Fish and Game), and Matthew Campbell (Idaho Fish and Game) for providing tissue samples for mitochondrial sequencing. Additionally, we would like to thank Dr. Masayuki Ushio, Dr. Nathan Griffiths, and one anonymous reviewer for their thoughtful feedback which helped improve the manuscript.