Using environmental DNA sampling to monitor the invasion of nonnative Esox lucius (northern pike) in the Columbia River basin, USA

This is an open access article under the terms of the Creat ive Commo ns Attri bution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2019 The Authors. Environmental DNA published by John Wiley & Sons Ltd 1United States Department of Agriculture, Forest Service, National Genomics Center for Wildlife and Fish Conservation, Rocky Mountain Research Station, Missoula, Montana 2Colville Confederated Tribes, Spokane, Washington


| INTRODUC TI ON
Aquatic invasive species are recognized as a global threat to conservation efforts for native species and create an economic cost to society (Gallardo, Clavero, Sánchez, & Vilà, 2016;Lovell, Stone, & Fernandez, 2006). Invasions of nonnative species can have adverse effects on property values, agricultural productivity, public utility operations, native fisheries, commerce, recreation, and ecosystem function (Mefford et al., 2017). In the United States alone, natural resources managers spend $120 billion annually repairing damage from and controlling the spread of invasive species (Pimentel, Zuniga, & Morrison, 2005). In the U.S. Pacific Northwest, the history, culture, recreation, and economy are tightly connected to the production of Pacific salmon and steelhead trout (Oncorhynchus spp.). As a result, resource managers and policy makers in the region are concerned about the impacts aquatic invasive species may have on the native fish community.
Since the mid-1900s, nonnative E. lucius have become established throughout the headwaters of the Columbia River basin, including western Montana and Idaho (McMahon & Bennet, 1996). The first documented sighting of E. lucius in Washington was in the Pend Oreille River in 2004 (Harvey, 2011). By 2010, E. lucius had spread downstream into the main stem of the Columbia River in Canada and the United States (Lee et al., 2010). In response, the Washington Fish and Wildlife Commission reclassified E. lucius as a Prohibited Species in 2011. This classification allowed unlimited harvest of E. lucius, required anglers to kill E. lucius before leaving the waterbody in which they were caught, and strictly prohibited the transport and release of live E. lucius into other waters (Washington Department of Fish and Wildlife, https ://wdfw.wa.gov/ais/esox_luciu s/). Although suppression efforts have reduced E. lucius abundance in the Pend Oreille River (Harvey & Bean, 2017), this species has continued to spread downstream into portions of Lake Roosevelt, a reservoir on the Columbia River in Washington (McLellan, Wolvert, Kittel, Silver, & Lee, in press Early detection of aquatic invasive species is critical for helping fisheries managers prevent further spread. However, detection with traditional sampling methods (e.g., electrofishing, seining, gillnetting, or angler surveys) over large geographic areas can quickly become cost prohibitive. In addition, sampling with traditional fisheries methods in remote areas can be difficult and time consuming. Under these circumstances, environmental DNA (eDNA) sampling may be a more effective tool for determining the presence and distribution of invasive species like E. lucius (e.g., Dejean et al., 2012;Herder et al., 2013;Jerde, Mahon, Chadderton, & Lodge, 2011;Wilcox et al., 2016 [appendix C & D]).
Two eDNA assays are currently published for detection of E. lucius. Olsen, Lewis, Massengill, Dunker, and Wenburg (2015), Olsen, Lewis, Massengill, Dunker, and Wenburg (2016) developed an assay in the mitochondrial cytochrome oxidase subunit I (COI) gene. This assay was intended for detection of invasive E. lucius in Alaska and was not extensively tested for use elsewhere. Spens et al. (2017) designed an assay based on a portion of the mitochondrial cytochrome b (cytb) gene. However, this assay was not central to the paper, which was primarily concerned with efficiencies in extraction and storage of DNA. Details concerning its development and testing of the assay are largely absent from the paper. As a result, the reliability of this assay is unknown.
Here, we describe the development of an eDNA assay to detect E. lucius throughout their introduced and native range in the United States, including Alaska and locations east of the Continental Divide.
Additionally, we evaluated the specificity of previously published assays by Olsen et al. (2015) and Spens et al. (2017) in silico to better understand the appropriate geographic range for each assay and to determine whether they could be paired with our assay for increased accuracy in eDNA results (e.g., Carim, Christianson, et al., 2016c).
Finally, we coupled this assay with eDNA sampling across the leading edge of the E. lucius invasion in the Columbia River basin. Our goals were to assess the efficacy of eDNA methods for detection of invasive E. lucius relative to the traditional methods and to gather additional information on E. lucius in areas around the active invasion in the Columbia River basin that are difficult to sample using traditional methods.

| Marker development and validation
We used publicly available data on GenBank to compile DNA sequences of the cytb gene of the mitochondrial genome for E. lucius (Table S1) and 15 other species whose ranges commonly overlap with E. lucius in western North America (Table S2). We excluded sequences of E. lucius from southern Europe in marker development as these fish may represent E. cisalpinus (Bianco & Delmastro, 2011; labeled as E. flaviae by Lucentini et al., 2011;Table S3) or other members of this genus (Denys, Dettai, Persat, Hautecoeur, & Keith, 2014). We used these sequences and the DECIPHER package (Wright, 2014) in R v. 3.0.1 (R Development Core Team, 2014 to develop a forward and reverse primer set that amplifies 108 bp of the cytb gene in the E. lucius mitochondrial genome: forward primer 5'-CAGCCACAATCCTCCATTTATTATTC-3'; reverse primer 5'-TGTAGGAGAAGTAGGGATGAAAGGG-3'. We designed a hydrolysis probe (5'-FAM-CCAGTAGGTAT TAACTCTGATG-MGBNFQ-3') by aligning target and nontarget sequences in MEGA 6.0 (Tamura, Stecher, Peterson, Filipski, & Kumar, 2013) and identifying a region specific to E. lucius. Sequences of E. masquinongy (muskellunge; the most closely related species to E. lucius in North America) have at least 3 bp mismatches in the probe and both primers (GenBank accessions AY497455AY497456; Table S3). We assessed the melting temperatures of the primers (forward: 59.9°C; reverse: 59.8°C) and probe (69.0°C; including minor groove binder moiety) in Primer Express 3.0.1 (Life Technologies).
The assay was tested using DNA extracted from 51 E. lucius tissues collected from 36 different locations across eight U.S. states, as well as 41 nontarget species (Table S4). We selected tissues from nontarget species that were either closely related (e.g., Esox spp.), or common in areas where E. lucius have been introduced. Tissue-de- We optimized primer concentrations following methods outlined in Wilcox et al. (2013) for final concentrations of 600 nM per reaction for both the forward and reverse primers (the probe concentration remained at 250 nM). We tested assay sensitivity and identified the limit of quantification by creating a six-level standard curve dilution series (6 250, 1 250, 250, 50, 10, and 2 copies per 4 µl). To create these standards, we cleaned PCR product from the above analysis (E. lucius tissue amplified with the assay) using the GeneJET PCR Purification Kit (Thermo Scientific) following the manufacturer's protocol. The standard curve was created by directly quantifying DNA using a Qubit 2.0 Fluorometer (Thermo Fisher Scientific) and then diluting the DNA in IDTE buffer (10 mM Tris, 0.1 mM EDTA, pH 8.0) to the desired concentrations.
We ran six replicates of each standard dilution using optimized primer concentrations and the cycling conditions above. We used StepOnePlus software v2.3 to estimate the efficiency and R 2 of the standard curve. To determine whether internal positive controls used to assess inhibition affected sensitivity of the assay, we repeated the standard curve analysis but replaced 1.8 μl of water in each reaction with TaqMan™ Exogenous Internal Positive Control (IPC; see below for more details).
To determine the limit of detection and verify accuracy of the standard curve dilutions, we followed methods above to dilute DNA to a concentration of 1 copy per 4 µl and analyzed this standard across 48 replicate reactions. Assuming a Poisson distribution, the probability of obtaining at least one copy of DNA in a reaction at this concentration is 0.63 (see Bustin et al., 2009). As a result, we would expect to detect DNA in about 63% of replicates (approximately 30 of 48).
We screened the assay in vivo using eDNA samples collected from 14 locations across 11 U.S. states where the presence or absence of E. lucius was known (Table S5). Samples were collected within 0.5 m of the water's surface by using a peristaltic pump to filter 5 L of water through a 47-mm diameter, 1.5-μm mesh glass microfiber filter (Whatman®) following methods described in Carim, McKelvey, Young, Wilcox, and Schwartz (2016a). Each sample filter was stored individually in silica desiccant at ambient temperature and shipped to National Genomics Center for Wildlife and Fish A sample was considered inhibited if the mean cycle threshold (Ct) for the internal positive control across the triplicate PCRs was delayed ≥ 1 cycle compared to the no-template control. We treated such samples with an inhibitor removal kit (Zymo Research) and reanalyzed each in triplicate. Removal of inhibitors may result in loss of DNA in a sample, but with elution volumes of 100-200 μl, loss of DNA during inhibitor removal is expected to be less than 10% on average (see http://www.zymor esear ch.com for more details). To minimize DNA loss during inhibitor removal, we extracted DNA from the second half of the sample filter and combined it with the extract from the first half to obtain ~200 μl of extracted eDNA prior to treatment. The sample was then reanalyzed following the methods above. If the reanalysis showed that a sample was still inhibited after treatment for inhibitor removal, we analyzed 12 µl of eDNA across four PCR replicates instead of three. Reaction volumes were maintained at 15 µl by increasing the volume of water to compensate for the reduced DNA volume in each of the four replicates. Diluting the amount of DNA in a given PCR by increasing the ratio of water to DNA can also reduce the effects of inhibitors during analysis (McKee, Spear, & Pierson, 2015).
Negative control samples were assessed at multiple stages throughout the protocol to identify any contamination leading to false positive detections. Negative controls were collected in the laboratory and analyzed to test kits containing sampling materials for contamination. One kit per 100 was used to collect a control sample by filtering 0.5 L of DI water as described above. To test for contamination during eDNA extraction, we processed an unused filter from every set of samples extracted (typically one extraction control for every 23 samples). Finally, a negative control was included on each PCR plate to ensure that PCR reagents and laboratory materials were not a source of contamination. used for in silico development of our assay to identify base-pair mismatches that may affect assay performance (Table S6). Because the Olsen et al. (2015) assay was developed in the COI region of the mitochondrial genome, this assay could not be compared to the reference sequences used in development of our assay. Instead, we screened this assay in silico by comparing it to 292 COI sequences of Esox spp.

| eDNA monitoring of E. lucius in the Columbia river basin
We applied our eDNA assay broadly throughout the upper Columbia River basin, from Lake Roosevelt downstream past Chief Joseph Dam, as well as major tributaries including the Kettle River, Colville River, Spokane River, Sanpoil River, and Okanogan River ( Figure 1a). This study area spanned the invasion, including areas where E. lucius are well established and where they are believed to be absent.
Extensive fisheries research and monitoring surveys have been conducted throughout Lake Roosevelt by its comanagers. This includes year-round angler surveys at 32 access points in Lake Roosevelt resulting in a total of 700 surveys a year (Blake, Kittel, & Nichols, 2017). In addition, the comanagers implement a reservoir-wide annual survey in the first week of November at 150 randomly selected sites using standard gill nets (Blake et al., 2017). Data collected from these surveys have been used to track the expansion of E. lucius throughout Lake Roosevelt since they were first detected in 2011. Due to limited funding and difficult access, little fisheries monitoring is conducted in the Kettle or Colville Rivers that drain into the upper reaches of the Lake Roosevelt. To avoid impacts to threatened and endangered fish species, gill-net surveys are rarely used in the main-stem Columbia River downstream of Grand Coulee Dam or in the Okanogan River.
Information from the traditional surveys was used to select 35 eDNA sampling sites. Of these, seven represented sites where E. lucius were known to be present and were collected in areas believed to support spawning and rearing. The remaining sites were in areas where E. lucius were believed to be absent (n = 20) or were near the invaded area where the presence of E. lucius was uncertain (n = 8; Table 1, Figure 1a, Table S7). At each site, we sampled areas with preferred E. lucius habitat (slow-moving water and emergent vegetation) and at the mouths of major tributaries to Lake Roosevelt. We treated sites with the same downstream extent but on opposite sides of Lake Roosevelt or tributary rivers as paired sites (n = 27 pairs, median distance between paired sites, 92 m, range 26-2104 m; Table S7). Only one sample per site was collected at locations in Banks Lake and tributary rivers with a wetted width of less than 5 m.  (Casselman & Lewis, 1996;Craig, 2008Pierce, 2012).

| Marker development and validation
Of the 71 E. lucius cytb reference sequences obtained from GenBank, eight contained a 1-bp mismatch with the E. lucius eDNA assay (Table   S1). The location of the mismatch varied, with three sequences F I G U R E 1 Locations of eDNA samples collected in the Columbia River basin for detection and monitoring of invasive Esox lucius. Numbers correspond to "Site ID" in Table 1. The Columbia River basin flows westward to the Pacific Ocean and is shown on the inset map in shaded yellow. Lake Roosevelt is a reservoir within the Columbia River, beginning at the border with Canada and extending to Grand Coulee Dam. (a) Observed distribution of E. lucius. Sites where live E. lucius had been observed prior to sample collection are labeled as "present" (red dots); sites near the invaded area where E. lucius presence is possible, but unconfirmed, are labeled as "possible" (yellow triangles); sites where E. lucius are believed to be absent are labeled as "not observed" (white squares). (b) eDNA detections of E. lucius. Sites with positive detections are labeled as "detected" (red dots); sites where eDNA of E. lucius was not detected are labeled as "not detected" (white squares). For sites with paired samples, a location is shown as positive if E. lucius DNA was detected in at least one of the two samples containing the mismatch in the forward or reverse primer and five sequences containing the mismatch in the probe. All sequences that were not an exact match to the E. lucius assay originated from fish collected in Sweden (n = 1) or Germany (n = 7). Sequences of the E. cisalpinus (Lucentini et al., 2011) exhibit up to 4 bp mismatches with the forward primer and up to 2 bp mismatches with the probe (Table S3).
DNA from all E. lucius tissue samples successfully amplified, and there was no amplification of nontarget species. Standard curve analysis resulted in an amplification efficiency of 93.9% (r 2 = 0.99, intercept = 38.66, slope = −3.48) and 100% detection across six replicates averaging two copies of DNA per reaction (mean Ct = 38.1).
When IPC was included in each replicate of the standard curve, the sensitivity of the assay was unaffected. Here, we observed amplification efficiency of 100.2% (r 2 = 0.99, intercept = 38.79, slope = −3.32) with 100% detection across six replicates with two copies of DNA per reaction (mean Ct = 37.9). Because IPC did not have an effect on the sensitivity of the assay, we did not include it in the PCR test for limit of detection at 1 copy per reaction. We detected E. lucius DNA in 33 of 48 PCR replicates with DNA template at 1 copy per reaction (Ct values, 36.6-39.8). These results indicate that our standard curve dilution was accurate and that the limit for detection and quantification of E. lucius DNA with our assay is 2 copies per reaction.
In vivo validation of the E. lucius eDNA assay was successful.
E. lucius DNA was detected in all samples collected where this species was known to be present (n = 2); E. lucius DNA was not detected in any samples collected in areas where this species was presumed absent (n = 12; Table S5).
In silico comparisons of E. lucius assays developed by Spens et al. (2017) and Olsen et al. (2015) indicated that they were potentially less species-specific. The Spens et al. (2017) assay was an exact match to 36 of 40 cytb sequences of E. cisalpinus in GenBank (Table S6). Similarly, the Olsen et al. (2015) assay was an exact match to 24 of 28 COI sequences of E. cisalpinus in GenBank (Table   S6) (Tables S1 and S6).

| eDNA monitoring of E. lucius in the Columbia river basin
E. lucius DNA was detected in 9 of 62 eDNA samples collected in the Columbia River basin (Table 1, Figure 1b). The nine positive detections were from the seven locations where live E. lucius had been previously observed during angler and gill-net surveys (Blake et al., 2017 and ). At two sites, E. lucius DNA was detected in only one of the paired bank samples (site 6a, the east bank of Lake Roosevelt near the town of Hunters, WA, and site 25a, the west bank of the Spokane River at the Lake Coeur d'Alene outlet). Additionally, a sample collected on the west bank of Lake Roosevelt near China Bend (site 4b) was positive for E. lucius DNA, but results for the sample from the east bank (site 4a) were inconclusive (see below). E. lucius eDNA was not detected in samples from 20 locations where this species was presumed absent or 8 locations where presence of this species was possible but unconfirmed.
Two samples showed signs of PCR inhibition: site 4a, at China Bend on the east bank of the Columbia River, and site 20, along the north bank of the Kettle River near Laurier (Table 1, Figure 1b).
After treatment to remove inhibitors, E. lucius eDNA was detected in the sample from site 20 with no further signs of PCR inhibition.
Inhibition was still present in the sample from site 4a after treatment. Subsequent dilution and reanalysis across four PCR replicates did not remove PCR inhibition from this sample, rendering the analysis inconclusive. There was no amplification of E. lucius DNA in any equipment or procedural negative controls.

| Assay performance
We have developed and rigorously tested an eDNA assay to assist managers in early detection and monitoring of invasive E. lucius in North America. The 71 cytb E. lucius sequences used in marker development represented individuals from throughout the Holarctic region, including Canada and Europe. Eight of these sequences from northern Europe were not identical to the eDNA assay. The assay developed by Olsen et al. (2015) may be the best choice for detecting E. lucius in northern Europe because its primer and probe are an exact match to all available E. lucius sequences from the area. It is, however, prudent to challenge an assay with local samples prior to application in areas where it was not formally tested. The presence of E. cisalpinus (Bianco & Delmastro, 2011) and a number of other described and cryptic taxa in the genus Esox (Denys et al., 2014;Skog, Vøllestad, Stenseth, Kasumyan, & Jakobsen, 2014) suggests caution in applying any of these assays to waters outside North America.
The sequences of all three assays are identical to all E. lucius reference sequences originating from North America, but the Olsen et al. (2015) assay contains no probe mismatches with E. masquinongy. Olsen et al. (2015) screened their assay in vitro against one sample of E. masquinongy of unknown origin. They did not screen E. masquinongy from multiple regions, or test whether higher concentrations of E. masquinongy DNA would result in nontarget amplification (see Wilcox et al., 2013). Our assay and that of Spens et al. (2017) have mismatches in both the primers and probes when compared to sequences of E. masquinongy DNA. As a result, these assays are a better choice for eDNA applications where E. lucius and E. masquinongy may co-occur. Of these two, the assay developed here has been tested against all North American Esox species, whereas the

Mean Ct
Banks Lake A limitation of this study is that we did not collect samples in a framework that allowed for robust modeling of detection probabilities at a given site. The probability of detecting a species with eDNA is related to both the abundance and proximity of individuals to the location of sample collection (Pilliod, Goldberg, Arkle, & Waits, 2013;Wilcox et al., 2016). Our initial work indicates that eDNA sampling is a viable strategy for monitoring the extent of E. lucius invasion in reservoirs. We believe that more increased temporal and spatial sampling throughout an area occupied by E. lucius could help describe seasonal movements and habitat preferences (e.g., Dunn, Priestley, Herraiz, Arnold, & Savolainen, 2017;Erickson et al., 2016;Erickson et al., 2017;de Souza, Godwin, Renshaw, & Larson, 2016). This information could help maximize suppression efforts by informing where and when they might be most effective. In addition, more extensive sampling outside the known distribution of E. lucius would promote a timely response to expansion of E. lucius in the Columbia River basin.

| eDNA strategies for invasive species
Understanding when to use eDNA methods and the best sampling design for detection of invasive species depends on research objectives and data needs. Survey methods that simultaneously detect invasive species and provide an opportunity for suppression (e.g., gillnetting) may be preferred when the leading edge of the invasion can be accurately estimated. This was the case in the upper portion of Lake Roosevelt where ongoing surveys and suppression efforts are underway. In contrast, eDNA methods may be a more efficient tool in areas that are monitored infrequently and where little information is known about the general fish community. Some eDNA researchers recommend including a negative field control for each field sample (Goldberg et al., 2016). Protocols for collecting this control sample typically include filtering a small amount of distilled or deionized water in the field. Such field controls are useful when evaluating field protocols and training field crews to collect eDNA samples. Paired negative field controls are imperative in protocols that require field decontamination for pieces of equipment that come into direct contact with every field sample (e.g., a van Dorn sampler). Yet, even these negative field controls will not test for contamination that results from error during field sample collection or from extraneous sources of DNA such as field crews or their equipment (see Merkes, McCalla, Jensen, Gaikowski, & Amberg, 2014;Song, Small, & Casman, 2017). Although we did not include negative field controls (sensu Goldberg et al., 2016), we assessed field contamination by including field samples from areas where the target species is known to be absent. Given the lack of northern pike DNA in these samples, we concluded that field contamination was not a factor influencing the results of this study.
Any field sample with surprising results (whether positive or negative) should be assessed with resampling at the same location (see also Dunker et al., 2016;Wilcox et al., 2016). In our study, suspect samples would be detections in areas distant from known occupied areas or detections in waterbodies representing poorquality habitat for E. lucius. Inconsistent detections over repeated sampling events may represent contamination (false positive detections), degradation, or very low quantities of DNA in the environment (false negative detections; e.g., Erickson et al., 2016), or movement of animals to or from adjacent areas (e.g., . Surprising detections should therefore be considered in the context of the larger dataset (e.g., proximity to the nearest known occupied habitat) as well as movement and habitat preferences associated with the target species' life history. Finally, the most appropriate interpretation of eDNA detections depends on study objectives. When monitoring invasive species, it may be more prudent to interpret positive eDNA detections as the presence of live individuals to ensure a proactive response to introduction and expansion of invasive species (Darling & Mahon, 2011).

| CON CLUS IONS
This study has demonstrated the importance of rigorous screening when developing an eDNA assay for broad-scale application.
Compared to other published assays, our assay appears to be the most specific and rigorously validated for detection of E. lucius in North America. Furthermore, our assay performs as well as traditional survey methods for coarse-scale monitoring of invasive E. lucius in the Columbia River basin. Increased spatial and temporal sampling is necessary to inform management on finer spatial scales and to allow for a rapid response to expansion of E. lucius in this system. Future monitoring should include sampling at all sites in both the spring and fall to better understand seasonal variation in detection based on life history, seasonal movements, and hydrology.

ACK N OWLED G M ENTS
We

DATA AVA I L A B I L I T Y
All data presented in this manuscript will be made publically available through the eDNAtlas https ://www.fs.fed.us/rm/boise/ AWAE/ proje cts/the-aquat ic-eDNAt las-proje ct.html and the National Genomics Center for Wildlife and Fish Conservation.