High stream flows dilute environmental DNA (eDNA) concentrations and reduce detectability

Environmental DNA (eDNA) is a rapidly emerging methodology with important applications to environmental management and conservation. However, the effects of stream flow or discharge on eDNA have been minimally investigated in lotic (stream and river) environments. In this study, we examined the role of stream flow on eDNA concentrations and detectability of an invasive clam (Corbicula fluminea), while also accounting for other abiotic and biotic variables.


| 1919
CURTIS Jerde et al., 2011), and may also be cost and time efficient relative to some standard sampling methodologies (Evans et al., 2017;Smart et al., 2016). However, uncertainties remain related to the physical, chemical, and biological factors influencing eDNA production, transport, and persistence in natural ecosystems (Barnes & Turner, 2016;Cristescu & Hebert, 2018). As such, substantial research is still needed to define the environmental and organismal contexts for which eDNA does and does not perform well.
Many eDNA studies have been conducted in lotic ecosystems (streams and rivers), where water flow directionality and variability through time (i.e. the hydrograph or flow regime; Poff et al., 1997) may strongly affect the performance of this method (Stoeckle et al., 2017). Yet the effects of stream flow on eDNA remain understudied (but see Akre et al., 2019;Jane et al., 2015;Shogren et al., 2018). High stream flows or floods may affect eDNA concentrations and the overall detectability of target organisms through multiple, potentially opposing mechanisms. High stream flows might be expected to dilute eDNA in the environment, thus reducing eDNA concentrations and potentially producing false negatives (Jane et al., 2015;Stoeckle et al., 2017). Similarly, transport of sediment or soil at high stream flows could cause false negatives through PCR inhibition Jane et al., 2015). Alternatively, high stream flows could re-suspend buried eDNA from stream or river sediments, consequently increasing detection probabilities and potentially leading to false positives (Turner et al., 2015). Further, at low flows it is expected that eDNA would be quickly removed from the water column as it settles onto the substrate, but during high flows eDNA may persist in the water column longer and be transported farther (Shogren et al., 2018). Researchers and practitioners working in environmental management need guidance on which of these opposing effects of stream flow on eDNA prevail under most conditions in order to design and implement appropriate sampling schemes.
Previous research on the effects of stream flow on eDNA has largely been restricted to artificial streams (Shogren et al., 2018), or short-duration enclosure (Jane et al., 2015) and tracer experiments (Fremier et al., 2019;Shogren et al., 2019). Results of some of these studies have been inconsistent with respect to the effects of stream flow or floods on eDNA concentrations and the detectability of target taxa. For example, Jane et al. (2015) enclosed Brook Trout (Salvelinus fontinalis) in two headwater streams, finding that during high flows eDNA copy number declined in one stream but increased in the other. Several experimental stream studies have provided mechanistic insight into how stream flow and substrate type affect eDNA transport and retention (Shogren et al., 2017(Shogren et al., , 2018, but it may be difficult to extrapolate from these small-scale, short-duration observations to the effects of large magnitude floods in natural ecosystems, as well as to estimate the relative importance of stream flow in comparison with other abiotic or biotic factors that can simultaneously affect eDNA. Specifically, is the effect of stream flow on eDNA concentration and detectability as important as the abundance or biomass of study organisms across disparate ecosystems (Yates et al., 2019), or as important as other abiotic factors like temperature or UV exposure that may influence eDNA production and persistence (Kessler et al., 2020;de Souza et al., 2016;Strickler et al., 2015)?
We propose that long-duration in situ studies are needed to generalize the effects of stream flow on eDNA concentrations and detectability, as well as to compare these effects concurrently to the role of other abiotic or biotic factors related to eDNA production, transport and persistence (Barnes & Turner, 2016). Here, we assessed how stream flow affects eDNA concentrations and detectability in situ using populations of an invasive freshwater mollusc, the Asian Clam (Corbicula fluminea). We used a longitudinal study to assess the role of stream flow, including high magnitude floods, on eDNA concentrations and detectability over an entire year at two stream sites, as well as a seasonal study (summer, autumn) to evaluate similar effects at eight stream sites over a gradient of low to high C. fluminea abundance. Together, our two studies provide the longest duration and largest-scale investigation of the effects of stream flow on eDNA concentrations and offer direct contrasts between the role of stream flow on eDNA relative to other abiotic or biotic factors. Our results should help researchers and practitioners design better eDNA sampling schemes for lotic environments and inspire subsequent studies on the relationship between stream flow and eDNA in order to replicate and validate our work in different ecosystems and taxonomic groups.

| Study species
Corbicula fluminea is a small (<5 cm) freshwater clam native to Africa, Asia and Australia that has invaded North America, South America and Europe likely through ballast water, bait releases, and intentional introductions as a food source (Crespo et al., 2015). Due to its rapid sexual maturity, high fecundity, and ability to reproduce both sexually and asexually (Hornbach, 1992), C. fluminea is considered among the most impactful aquatic invasive species globally (Sousa et al., 2008). Invasions of C. fluminea have been reported to alter biogeochemical cycling (Turek & Hollein, 2015), negatively affect native mollusc species (Haag, 2019), clog water intakes and canals (Isom, 1986), and cost over $1 billion/year in the United States (US) to manage (Pimentel et al., 2005). We chose C. fluminea as our focal species due to its prevalence in our study sites, the availability of a genus-specific eDNA assay for Corbicula (Cowart et al., 2018), and our hope that these results will be relevant to the management of this invader. While multiple forms and species of Corbicula have invaded the USA (Haponski & O' Foighil, 2019;Tiemann et al., 2017), only C. fluminea was found at our study sites.

| Longitudinal eDNA study
We selected two streams in central Illinois (Champaign County, US) equipped with US Geological Survey (USGS) flow gages that measure stream flow every 15 min. Additionally, these streams represented a contrast in stream flow and density of C. fluminea. Copper Slough is an urbanized, flashy, headwater stream (12.8 km 2 drainage area) with gravel-sand substrate that drains into the Kaskaskia River (Table 1; Figure S1). Salt Fork is a larger (215.7 km 2 drainage area) rural stream with sand substrate, surrounded by row-crop agricultural fields, that drains into the Vermillion River of the Wabash River watershed (Table 1; Figure S1). We planned to sample Copper Slough bimonthly and Salt Fork monthly (Figure 1). In addition to this planned sampling, we opportunistically sampled low-flow and high-flow events, including before rain and during rising and falling limbs of the hydrograph. We sampled these two stream sites from 11 January to 27 December 2018. In total, Copper Slough was sampled 33 times and Salt Fork was sampled 24 times (Figure 1).
At each sampling event, we collected four 250 ml surface water samples across the width of each stream at the location of the USGS flow gage. During high-flow events, when wading across the stream was not possible, we collected water samples using buckets lowered from the bridge at the location of the USGS flow gage. All bottles had been washed with 50% bleach prior to use. Buckets and bottles were triple-rinsed with stream water at the site prior to water collection.
After sample collection, bottles were sealed in a clean plastic bag, placed on ice in a cooler and filtered <2 hr after collection. During one collection, samples from Copper Slough were refrigerated and filtered the following morning (~12 hr after collection) but this delay does not affect the quantity of eDNA recovered (Curtis et al., 2021).
For each sampling event, one field blank of distilled water per site was used to assess potential contamination in collection supplies.
During each water collection (after water samples were collected), water quality parameters, including pH, temperature, salinity and total dissolved solids (TDS), were recorded using a handheld probe (Oakton ® ) and turbidity was recorded using a portable metre (Sper Scientific © ; Table S1). We wore nitrile gloves to collect all water samples, filter water samples, and during all laboratory procedures, with frequent glove changes.
Water samples were transported to the University of Illinois at Urbana-Champaign (UIUC), where we cleaned the bench space with a 50% bleach solution prior to filtration and used supplies (funnels, forceps) that had been previously washed with 50% bleach (Goldberg et al., 2016). We then vacuum filtered samples onto 0.8 µm cellulose nitrate filters (Whatman™, General Electric Healthcare) and submerged the filters in 900 µl of cetyl trimethylammonium bromide (CTAB) in a 2 ml microcentrifuge tube (Renshaw et al., 2015). We kept these tubes in the dark at room temperature for 1 month to increase cell lysis (Wegleitner et al., 2015) and then placed them into a −80°C freezer until extraction.

| Seasonal eDNA study
We examined the relationship between C. fluminea density and eDNA concentration and detectability, as well as the dependency Corbicula fluminea density was quantified adjacent to USGS flow gages by sampling 15 random quadrats (0.25 m 2 ) within a stream length that was ten times its wetted width, following a modified version of systematic random sampling for freshwater mussels in the US (Strayer & Smith, 2003). We reduced sampling effort from the 40 recommended quadrat samples of Strayer and Smith (2003)  Environmental DNA sample collection for the seasonal study followed the same methods as above for the longitudinal study. We collected four 250ml surface water samples per site and recorded water quality parameters (Table S2). We used one field blank (distilled water) per site to assess background contamination in bottles and filtering supplies. We bagged water samples, placed them on ice, transported them to UIUC, and filtered and stored samples consistent with our preceding methods. We collected summer eDNA water samples prior to conventional C. fluminea density sampling at every site to minimize risk of contamination that might arise from handling our study organism.

| Corbicula eDNA assay
Initially, we used a Corbicula genus-specific assay with the following primers developed by Cowart et al. (2018) to amplify a 208 bp region of the COI gene: where C. fluminea amplification was determined in this primer-only assay by a melt curve temperature of ~76.61°C. However, we noticed in March 2018 that two different peaks were present in the melt curve analysis of some samples. One peak was consistent with C. fluminea but the other showed amplification of a non-target organism at a melt curve temperature of ~85.15°C ( Figure S2). We used Sanger sequencing to determine that this non-target amplification was a F I G U R E 1 Stream flow (blue hydrograph) and temperature (black diamonds) at each sampling event (grey lines) during the longitudinal study ofC. flumineaeDNA at Salt Fork (a) and Copper Slough (b), Champaign County, Illinois, US, during the 2018 calendar year gut bacterium (Klebsiella spp.) that we postulate was present due to abundant Canada Goose (Branta canadensis) populations at our sites.
The dual amplification of C. fluminea and Klebsiella spp. affected our ability to quantify copy number, so we developed a genus-specific probe (5′-FAM-AGTGATGCCAATAATAATGGGTGGTTTTGG-MGB-NFQ −3′) to eliminate this non-target amplification. Subsequent sequence confirmation indicated that only C. fluminea amplified with this primer-probe assay. We then ran or re-ran all eDNA samples in our study with the new genus-specific probe-based assay. Like the assay developed by Cowart et al., (2018), our assay cannot discriminate between some congeners within the Corbicula genus, all of which are non-native to North America, including undescribed cryptic species or forms (Haponski & Ó Foighil, 2019;Tiemann et al., 2017). Prior to use of this assay, we ran optimization of different primer and probe concentrations, with different annealing temperatures, and selected the combination that produced the earliest Cq values and highest efficiency (%) of serial dilutions. We report specifics on assay performance below.

| eDNA extraction and qPCR
Prior to extractions or qPCR preparation, the laboratory space was cleaned with a 50% bleach solution and UV treated for 20 min.
Following Renshaw et al. (2015), we extracted DNA from filters using a chloroform-isoamyl alcohol extraction procedure in a clean room, free from high-copy DNA and isolated from the PCR laboratory. This extraction method (Renshaw et al., 2015) has been shown to be robust to inhibition from tannins or humic acids (Curtis & Larson, 2020;Schrader et al., 2012) and often produces higher DNA yields than other extraction procedures (Deiner et al., 2015). One extraction blank was used for every ~25 samples.
Quantitative PCR reactions used the following: 10 µl TaqMan Environmental Master Mix 2.0 (Applied Biosystems ® ), 6.15 µl of sterile water, 0.35 µl of each primer (10 µM), 0.15 µl probe (10 µM) and 3 µl of eDNA. Negative plate controls replaced 3 µl of DNA with 3 µl of the master mix (Cowart et al., 2018). Plates were prepared in an isolated, PCR product-free clean room, then ran on a QuantStudio 3 Real-Time PCR system (Applied Biosystems ® ) using the following qPCR parameters: 95°C for 10 min denaturation and 40 cycles at 95°C for 15 s and 62°C at 1 min. We used a synthetic COI gBlock fragment (Integrated DNA Technologies) with GenBank accession GQ401362 (base 44 to 543) to develop serial dilutions (1:10) typically from 4.5 × 10 6 copies/µl (1 × 10 -3 ng/µl) to 4.5 copies/µl (1 × 10 -9 ng/µl), which we used to create a standard curve (Cowart et al., 2018). We ran all samples in triplicate and considered amplification in 1/3 plate replicates as a positive detection. To confirm that positive amplifications were Corbicula, one randomly selected sample per plate was cleaned with ExoSap-It Express™ (Applied Biosystems™) and Sanger sequenced at the University of Illinois' W.M. Keck Center. We edited sequences in Geneious © to remove ambiguities, realigned, and then using NCBI's BLAST confirmed that all sequences were C. fluminea.
No field blanks or negative plate controls amplified. Serial dilutions of positive C. fluminea DNA produced R 2 values that ranged between 0.991 and 0.999 and efficiencies between 91% and 100%.
To calculate copy number in eDNA samples, we used the Thermo Fisher Scientific DNA copy number calculator to determine the number of copies/ng (4,454,142,012 copies/ng for C. fluminea) and multiplied that by the quantity calculated for each well replicate, based on the standard curve. We defined our limit of detection (LOD) as the lowest standard that amplified in 1/3 of replicates and our limit of quantification (LOQ) as the lowest standard that always amplified in 3/3 replicates. Here, our LOQ was 4.5 copies/µl (~1 × 10 -9 ng/µl) and our LOD was 0.45 copies/µl (~1 × 10 -10 ng/µl).

| Statistical analyses
For the longitudinal study, we analysed eDNA copy number and omitted detection probability modelling (below; seasonal study) because C. fluminea was always present in these two study streams and eDNA was almost always detected (56 of 57 sampling events had positive detections). We averaged plate and site replicates to produce one eDNA copy number estimate for each sampling event per site. We then related eDNA copy number to stream flow and a series of additional predictor variables, including temperature, number of daylight hours, and Julian day, via multiple regression. Beyond our primary focus on stream flow, we included temperature because cooler temperatures may allow eDNA to persist longer in the environment (Eichmiller et al., 2016), whereas warmer temperatures are associated with increased reproductive activity of C. fluminea, particularly in the late spring or early summer (Hornbach, 1992).
Such reproductive events have been found to have strong effects on eDNA concentration and detectability in other study systems (e.g. de Souza et al., 2016).
We included daylight hours as a measure of potential UV exposure, which can influence eDNA degradation (Kessler et al., 2020;Strickler et al., 2015), and Julian day of the year to control for seasonal variation in C. fluminea eDNA production, persistence, or transport independent of our other predictor variables. Daylight hours and Julian day were obtained from the US Naval Observatory Astronomical Applications Department. We omitted other abiotic measures of stream water chemistry from our models because these generally varied little over time or between our stream sites (Table S1, S2). Streams in our study region have a slightly basic pH with relatively high conductivity, TDS, salinity and turbidity values owing to both the underlying sedimentary geology of central Illinois and intensive human land use in these watersheds (Larimore & Bayley, 1996;Walser & Bart, 1999). Turbidity did vary predictably with stream flow (i.e. higher turbidity during floods), but given the extent to which these prospective predictor variables were confounded with each other, we opted to use stream flow as the primary predictor in our models.
To better meet assumptions of linear regression, we log + 1-transformed eDNA copy number and stream flow. We then fit linear mixed-effects models using the package "lme4" (Bates et al., 2015) in R (v. 4.0; R Core Team, 2020) to examine relationships between eDNA copy number and all combinations of predictor variables (stream flow, temperature, daylight hours, Julian day), as well as a null model with no predictors. We used mixed-effects models with a random effect for site to account for within-site replication at Copper Slough and Salt Fork. We compared competing regression models of C. fluminea eDNA copy number using the Bayesian information criterion (BIC) under an information theoretic approach, in which model fit is evaluated while penalizing for model complexity (Burnham & Anderson, 2002). We used BIC rather than Akaike's information criterion (AIC) because AIC may lead to over-fitting in multi-model comparisons (Dennis et al., 2019). We calculated BIC using the "MuMIn" package (Bartoń, 2020) in R (v. 4.0; R Core Team, 2020). The most supported model in this approach has the lowest BIC value.
For the seasonal study, we first followed a similar statistical approach as in the longitudinal study, in which we sought to explain log + 1-transformed C. fluminea eDNA copy number by several predictor variables. Exploratory data analyses revealed that temperature and stream flow were highly correlated between our summer and autumn sampling events (Pearson's r = −0.70; Figure 2), and consequently, we combined these predictors into a single variable for season (summer or autumn). We also included summer C. fluminea density at our sites as a predictor of eDNA copy number, specifically seeking to compare the role of season (and its associated differences in stream flow) to density in explaining eDNA concentrations. We again fit linear mixed-effects models in the package "lme4" with site as a random effect and compared model performance with BIC calculated using the "MuMIn" package (Bartoń, 2020) in R (v. 4.0; R Core Team, 2020).
Next, we investigated the role of season and C. fluminea density on eDNA detectability, because we had more frequent non-detections of C. fluminea eDNA in the seasonal than longitudinal study.
We used occupancy modelling, a hierarchical regression approach common in fisheries and wildlife research, which estimates both the probability that a site is occupied by a species (occupancy, Ψ) and the probability of detecting the species when present (detection, p) using repeated observations over either space or time (MacKenzie et al., 2002). Environmental DNA sampling is amenable to this occupancy modelling framework because replicated field or laboratory samples can be used as the unit of response for estimating both occupancy and detection probability of the eDNA of the focal organism (Dorazio & Erickson, 2017;Schmidt et al., 2013). We conducted occupancy modelling using the "unmarked" package (Chandler et al., 2020) in R (v. 4.0; R Core Team, 2020), where the presence or absence of C. fluminea eDNA in our four field replicates per season and site was the response variable. We modelled occupancy of C. fluminea eDNA using our summer density estimate of this species, anticipating that the eDNA of this species was more likely to be present at higher rather than lower abundances. We modelled the detection probability of C. fluminea eDNA using both the summer density estimate and season. We expected that detection probability should increase with increasing abundance, and that the high stream flows associated with our autumn sampling could affect eDNA detection probability by the type of opposing dilution or transportation/mobilization effects explained in our introduction. We compared these competing occupancy models with BIC calculated using the "MuMIn" package (Bartoń, 2020)

| Longitudinal eDNA study
Two competing models for C. fluminea eDNA copy number were similarly supported (ΔBIC ≤2; Table 2). Our most supported model included stream flow and temperature as predictors, with a BIC weight  Figure 3b). Two competing models for C. fluminea eDNA copy number were equally supported (ΔBIC ≤2), which included season and C. fluminea density (BIC wi = 0.53; adjusted r 2 = .59; Table 3) and only season (BIC wi = 0.43; adjusted r 2 = .47). Season had a significant effect on eDNA copy number (t-value = 3.35, p = .01) with higher copy numbers observed in the summer. Conversely, we found a positive but non-significant relationship between C. fluminea density and eDNA copy number (t-value = 1.71, p = .14; Figure 4). Our most supported model for C. fluminea eDNA occupancy and detection probability included season and C. fluminea density as detection covariates and no covariates for occupancy (BIC w i = 0.62;

F I G U R E 2
adjusted r 2 = .60; Table 4). A competing model with only season as a detection covariate was equivalent by ΔBIC (BIC wi = 0.35; adjusted r 2 = .48). Per the most supported of these two equivalent models, we found a significant effect of season on C. fluminea eDNA detectability (z-value = 3.81, p < .001), and a positive but non-significant effect of density on C. fluminea eDNA detectability (z-value = 1.83, p = .07).
We used these results from our overall most supported model to estimate the number of water samples necessary to achieve a 95% cumulative probability of detecting C. fluminea eDNA in both summer and autumn over a density gradient for this species (Figure 5a-d).
Corbicula fluminea eDNA is relatively easy to detect in the summer, requiring three water samples at low densities (<15 individuals/m 2 ;

| D ISCUSS I ON
We found that high stream flows diluted eDNA concentrations and decreased detectability for our target organism, the Asian Clam C. fluminea. Our year-long longitudinal study at two sites revealed that this dilution effect of higher stream flows was significant, but weaker than the strong positive effect of temperature on C. fluminea eDNA concentrations and detectability, which we believe is largely attributable to late spring or early summer reproduction by this species (Aldridge & McMahon, 1978). Conversely, our seasonal study at eight sites revealed that the combined effect of both temperature and stream flow was stronger than the relatively weak and non-significant positive effect of C. fluminea density on eDNA concentrations and detectability. This is consistent with emerging consensus that organismal abundance or biomass is often positively but weakly associated with in situ eDNA concentrations or detectability (Yates et al., 2019), and our study supports that other abiotic (e.g. floods) and biotic (e.g. reproductive timing) factors may be as or more important than the density of focal organisms in explaining eDNA results. Cumulatively, our study reveals that eDNA applications for environmental management and natural resource conservation can be highly sensitive to the abiotic and biotic context of field sampling. If using eDNA to monitor for presence of invasive C. fluminea populations in lotic ecosystems, we would recommend TA B L E 2 All candidate models for eDNA copy number from the longitudinal eDNA sampling using linear mixed-effect modelling with site as a random effect Note: BIC is Bayesian information criterion; ΔBIC is the difference between the lowest BIC value and the model; BIC w i is the weight of the model compared to all models; and adj r 2 is the likelihood-based pseudo r 2 values used to assess model fit.
sampling at low or base stream flows in warm conditions when this species may be reproductively active (late spring and early summer), and we would recommend against opportunistic or ad hoc convenience sampling during high stream flows or floods. Resource managers and practitioners seeking to use eDNA to monitor populations or communities of interest will likely need to optimize their sampling designs to account for the variable natural history of study organisms and their ecosystems.

F I G U R E 3 Predicted relationships,
with 95% confidence intervals, from the most supported model (Table 2)  A future priority for studies of stream flow effects on eDNA may be to incorporate more information on stream geomorphology, including underlying soils or geology .
Differences in stream substrate have been shown to influence transport and resuspension dynamics of eDNA in artificial streams and flow-through columns, where finer sediments (i.e. sand, pea gravel) retain eDNA and result in shorter transport distance relative to larger substrates (Shogren et al., 2016(Shogren et al., , 2017. Similarly, during rain events, surface runoff flushes terrestrial soil and organic matter into streams, which could potentially bind with eDNA and reduce detectability (van Bochove et al., 2020). Clay soils in particular form a tight bond with DNA and can result in lower yield extractions of eDNA (Yankson & Steck, 2009), although studies of pond sediments have found weak and inconsistent effects of clay on eDNA detection probabilities Buxton et al., 2018). Soils at our eight study streams are classified as relatively similar silty clay loams or silt loams (USDA-NRCS, n.d.), and future studies might investigate whether stream flow effects on eDNA concentrations and detectability are similar for streams of different underlying soil types or geologies . Note: K is the number of parameters; BIC is Bayesian information criterion; ΔBIC is the difference between the lowest BIC value and the model; BIC w i is the weight of the model compared to all models; and adj r 2 is the likelihood-based pseudo r 2 values used to assess model fit. Note: BIC is Bayesian information criterion; ΔBIC is the difference between the lowest BIC value and the model; BIC w i is the weight of the model compared to all models; and adj r 2 is the likelihood-based pseudo r 2 values used to assess model fit.

F I G U R E 4 Model predicted relationship (
We found a strong positive effect of temperature on eDNA concentrations and detectability for C. fluminea. In past eDNA studies, temperature has often been associated with degradation, as warmer temperatures accelerate the breakdown of eDNA (e.g. Strickler et al., 2015). Alternatively, some in situ studies have instead found increased organismal activity, metabolism, or reproduction associated with warmer water temperatures to increase eDNA concentrations or detectability de Souza et al., 2016;Wacker et al., 2019). We believe that our positive association between Slough and Salt Fork, and as such, we believe the strong positive F I G U R E 5 Relationship between sampling effort (x-axis) and cumulative detection probability ofC. flumineaeDNA (y-axis) for autumn and summer seasons with 95% confidence intervals from the most supported model (Table 4)  reproduction. Past studies have emphasized that timing of reproductive events can be used to inform or optimize eDNA sampling for target organisms (de Souza et al., 2016;Spear et al., 2015), and we similarly would identify late spring or early summer temperatures approaching 25°C to be ideal for eDNA monitoring of C. fluminea populations.
In our seasonal study, we found weak positive but non-signif- high field replication as necessary to detect this organism in autumn even when abundant. As such, our study supports past work that has found season of eDNA sampling to strongly affect performance of this method (Buxton, Groombridge, Zakaria, et al., 2017;Spear et al., 2015;Wacker et al., 2019). Managers or practitioners should tailor eDNA monitoring programmes to the seasonal natural history of their study organisms and associated environmental conditions.
Given that organisms may differ dramatically in their detectability between seasons (i.e. de Souza et al., 2016), the design of multitaxa eDNA surveys may be particularly challenging, as the best time of the year to sample for one priority taxa may not be optimal for another.
Our results identify several understudied but critical considerations for future eDNA applications in lotic ecosystems, but our findings may also be applicable to marine systems, where similar dilution and transport of eDNA from currents has been documented (e.g. Andruszkiewicz et al., 2019). First, high stream flows can dilute eDNA concentrations and produce false negatives, even in cases where study organisms are relatively abundant. We recommend that researchers and managers or practitioners avoid eDNA sampling during high stream flows or floods. It may not be feasible or biologically relevant to always collect eDNA samples during low stream flows. For example, some target species may be most active or reproduce during higher flows of the autumn or winter (de Souza et al., 2016). Accordingly, if researchers must take eDNA samples during periods of high stream flows or floods, we recommend increased sample replication to improve detection probabilities (Sepulveda, Schabacker, et al., 2019).
Second, understanding the natural history of target species, and relationships to seasonal variability in abiotic and biotic conditions, should lead to improved eDNA sampling programmes. In our longitudinal study, apparent reproduction by C. fluminea in late spring and early summer resulted in high eDNA copy numbers at this time of the year, and monitoring programmes for this invasive species might seek to sample at temperatures associated with reproduction to improve detection probabilities.
Alternatively, studies seeking to relate adult C. fluminea densities to eDNA concentrations might instead strive to avoid this pulse of veliger-associated eDNA. Lastly, occupancy modelling is a useful tool for quantifying detection probabilities in eDNA studies, and associated power analyses can provide guidance on sampling effort necessary to detect target organisms under environmental conditions (Dorazio & Erickson, 2017;Schmidt et al., 2013).
Researchers and managers should continue to apply occupancy estimation with detection probability frameworks to improve the design and implementation of eDNA sampling programmes for specific taxa and ecosystems, including for lotic ecosystems where variable stream flow and floods may strongly affect performance of this methodology.

This research was funded by US Department of Agriculture National
Institute of Food and Agriculture Hatch Project ILLU-875-976