Introduction

The characteristic lag time between introduction and widespread establishment of biological invaders provides a window of opportunity for early detection and eradication of potentially harmful species in novel environments (Myers et al. 2000; Crooks 2005; Lodge et al. 2006; Mehta et al. 2007). Environmental DNA (eDNA) surveillance is a method for improving early detection efforts for rare aquatic species, including harmful species at the leading edge of an invasion front (reviewed by Rees et al. 2014). Environmental DNA (eDNA) surveillance is especially well suited for aquatic environments because cells and sloughed tissues are suspended in water and can be collected and screened to detect DNA of target organisms that are present but difficult to detect with conventional surveillance tools (Goldberg et al. 2011; Jerde et al. 2011; Sweeney et al. 2011; Dejean et al. 2012; Pilliod et al. 2013; Takahara et al. 2013; Jane 2014). Early applications of eDNA for aquatic invasive species (AIS) surveillance utilized a traditional endpoint polymerase chain reaction (PCR) approach, but eDNA methods are rapidly evolving and recent advances in the development of quantitative PCR (qPCR) assays for sample screening highlight the potential for improved sensitivity of eDNA based surveillance (Beja-Pereira et al. 2009; Bott et al. 2010; Thomsen et al. 2012a, b; Wilcox et al. 2013; Nathan et al. 2014; Turner et al. 2014a, b). Here we describe the application of a qPCR screening assay with a case study for surveillance of Ruffe, a federally designated ‘nuisance species,’ across the Laurentian Great Lakes (Nonindigenous Aquatic Nuisance Prevention and Control Act of 1990, Public Law 101–646).

Ruffe, a percid fish native to western and northern Europe, was first observed in North America in the St. Louis River (the westernmost tributary of Lake Superior) in 1986 (Pratt 1988), having likely been introduced via ballast water of vessel(s) originating from a port associated with the Elbe River drainage (Stepien et al. 2005). A rapidly reproducing population of Ruffe in the St. Louis River, combined with declines in some native fish populations and a persistent eastward advance, led the Aquatic Nuisance Species Task Force, under the authority of the Nonindigenous Aquatic Nuisance Prevention and Control Act, to declare Ruffe a ‘nuisance species’ in the spring of 1992. This designation triggered formal surveillance efforts by the US Fish and Wildlife Service (USFWS) and the Ontario Ministry of Natural Resources to detect pioneering populations of Ruffe in the Great Lakes. Ruffe is now known to inhabit many tributaries along the US shoreline of Lake Superior, from Duluth, Minnesota to the Tahquamenon River, Michigan. Range expansion into Lake Huron and northern Lake Michigan has also been documented and is presumed to have occurred as the result of larval fish transport in the ballast water of bulk carrier vessels traveling from invaded ports in Lake Superior or abroad (Bowen and Keppner 2013). The most recent USFWS surveillance reports indicated presence of Ruffe in the Cheboygan River (Cheboygan, MI, Lake Huron) and in Green Bay (near Escanaba, MI and Marinette, WI, Lake Michigan; Bowen and Keppner 2013). To date, live Ruffe have not been observed in Lake Erie or the southern basin of Lake Michigan, where it could potentially bridge the hydrological divide that separated the Great Lakes basin from the Mississippi River basin before the construction of the Chicago Sanitary and Ship Canal (US Army Corps of Engineers 2014).

Using current Ruffe distribution data from USFWS and the US Geological Survey Nonindigenous Aquatic Species database (USGS-NAS; http://nas.er.usgs.gov/default.aspx), along with predictions of future Ruffe distribution derived from a recently developed model of ballast-mediated Ruffe dispersal in the US waters of the Great Lakes (Sieracki et al. 2014), we designed an eDNA surveillance study to target Ruffe at the putative leading edge of the invasion. The main objectives of our surveillance effort were to: (1) monitor invasion of high-risk ballast uptake ports in the upper Great Lakes that could seed Lake Erie or the southern basin of Lake Michigan; (2) assess spread of Ruffe in eastern Lake Superior and in the northern parts of lakes Michigan and Huron; and, (3) search for potential incursions of Ruffe in Lake Erie and southern Lake Michigan. Our findings suggest a far more advanced invasion front for Ruffe in the Great Lakes than has been documented using the conventional fisheries methods employed for formal Ruffe surveillance efforts. We propose a conceptual framework to facilitate application of our results for the management of Ruffe, based on axes that reflect the weight of eDNA evidence of species presence and the likelihood of secondary spread.

Methods

Marker development

Molecular markers for G. cernua were designed using publically available sequence information (GenBank, www.ncbi.nlm.nih.gov). We assembled sequences for 23 Percidae species (22 of which are historically found in the Great Lakes basin plus Ruffe) for the cytochrome b, cytochrome c oxidase subunit 1 (COI), and control region as these regions of the mitochondrial genome were best represented in GenBank for the taxa of interest. Three COI primer pairs and two primer pairs from the control region were evaluated with tissue-extracted DNA from G. cernua and 17 of the co-occuring Great Lakes Percidae species. DNA extractions were all normalized to 1 ng/µL and tested under the qPCR conditions described below (see qPCR amplification and evaluation). Amplifications were post-PCR cleaned with ExoSAP-IT (USB) and submitted to the Genomics and Bioinformatics Core Facility at the University of Notre Dame (GBCF) for unidirectional Sanger Sequencing. Sequences were visually inspected with Sequencher™ (GeneCodes) and submitted to BLAST on NCBI. The optimal pair of primers, Ruffe_COI_Fa (5′-TACCCTCCCCTATCAGGAAACTT-3′) and Ruffe_COI_Ra (5′-TAATTGCGCCCAAGATTGAGGAGAT-3′), targeted a 111 bp fragment of the COI.

DNA extracted from tissues of several nontarget taxa (Etheostoma caeruleum, E. blennoides, E. zonale, Percina caprodes, and P. copelandi) produced amplicons that were identified as G. cernua by Sanger sequencing. As the tissue samples were provided by a museum collection, the potential for contaminating DNA (G. cernua) in the samples seemed plausible and the ability of the assay to identify the trace contaminant DNA in an environment dominated by the DNA of the non-target taxon added support for the stringent application of the assay on eDNA samples. Additional details for the design and validation of the assay are given in “Appendix 1: Detailed description of methods for qPCR marker development” section.

Sample collection, filtration, and extraction

From October 25, 2012 to September 11, 2013, we collected 1289 2-L water samples from 24 discrete locations in the Great Lakes basin (Fig. 1). We focused our sampling efforts on ‘high-risk’ locations at the leading edge of the known Ruffe invasion front (e.g. eastern Lake Superior, northern Lake Huron from Alpena to the Cheboygan River, and Green Bay) and on sites receiving the largest volume of shipping traffic from Ruffe infested ports within the US waters of the Great Lakes (e.g. Chicago, IL and Toledo, OH; National Ballast Information Clearinghouse 2014). Surface-water samples were collected in autoclaved 2-L Nalgene bottles (45 min at 121 °C). Subsurface water samples were collected within one meter of the bottom with a Van Dorn sampler (2.2 L opaque PVC; Wildlife Supply Co., Yulee, FL) or Kemmerer sampler (2.2 L acrylic; Wildlife Supply Co., Yulee, FL) and transferred in the field to autoclaved 2-L Nalgene bottles. All sample locations were geographically referenced with GPS (Garmin Dakota 10; s.e., <10 m). Samples were filtered and extracted as recommended by Mahon et al. (2010). Briefly, samples were vacuum filtered onto 1.5 µm pore-size glass fiber filters within 24 h of collection, filter papers were stored at −20 °C, and DNA was extracted with the PowerWater DNA Isolation kit (MO-Bio Laboratories Inc., Carlsbad, CA).

Fig. 1
figure 1

Sampling locations for 2012–2013 Ruffe eDNA surveillance effort

Full size image

Apart from the 2-L Nalgene bottles, which were autoclaved as described above, all equipment used in the sampling and screening effort, including boats, was sterilized with a 10 % bleach solution or sourced directly from suppliers (e.g. latex gloves). Cooler blanks, a single 2-L bottle filled with deionized water, were placed in each sample cooler and taken into the field. The cooler blanks were opened in the field, resealed, and then submerged into the waterbody being sampled. Prior to filtering each sample, approximately 500 mL of deionized water was passed through each sterilized filter apparatus onto a filter paper to test for contamination on lab equipment; these samples are referred to as equipment controls. All cooler blanks were screened for contamination and, for every field sample that tested positive, the corresponding equipment control was processed.

qPCR amplification and evaluation

qPCR amplifications were conducted using a SYBR® Green I dye assay. The qPCR amplification cocktail consisted of 1X Power SYBR® Green Master Mix (Life Technologies), 300 nM of each primer, 0.4 µg/µL of Bovine Serum Albumin (Ambion), and 4 µL of extracted DNA in a 20 µL reaction. We performed all reactions on an Eppendorf Mastercycler ep realplex 2 thermocycler. Thermal cycling conditions were as follows: an initial activation step at 95 °C for 10 min; 40 cycles of 95 °C for 15 s followed by 60 °C for 1 min; and, a melting curve analysis transitioning from 60 to 95 °C over 20 min. The fluorescence threshold for each plate was automatically determined by the Eppendorf realplex software using the default Noiseband setting. The fluorescence baseline was calculated for every reaction individually using the default Automatic Baseline setting of the Eppendorf realplex software.

Each eDNA extract was run in triplicate with a single positive control (tissue-derived DNA) and single negative control (1X TE buffer, low EDTA) included on each qPCR plate. Every amplification profile and melt curve profile was visually examined to confirm exponential amplification and a melting temperature matching that of tissue derived reactions. Successful amplifications were post-PCR cleaned with ExoSAP-IT (USB) and submitted to the GBCF for unidirectional Sanger Sequencing. Sequences were checked visually with 4 Peaks (http://nucleobytes.com/index.php/4peaks) or Sequencher™ (GeneCodes) and specificity to Ruffe was verified by BLAST on NCBI. eDNA extracts were considered positive for the presence of Ruffe DNA only after successful amplification and successful sequence confirmation. Final confirmation for a positive sample required a negative result from the corresponding equipment control.

Assessment of failure to detect target DNA

To assess to what extent our sample screening protocol (i.e. eDNA extract run in triplicate) failed to detect target species DNA when it was present in a sample, we screened samples collected in 2013 a second time. All samples that tested negative in the first round of screening were re-assayed in sextuplet (i.e. 6 technical qPCR replicates) with the appropriate controls in place as in previous analyses. As before, successful amplifications were post-PCR cleaned and submitted for unidirectional Sanger sequencing and equipment controls were screened for final confirmation of a positive DNA detection.

Results

A total of 72 samples across fourteen locations tested positive for Ruffe DNA (Fig. 1). All positive detections were from samples collected over the period May to July 2013 (Table 1). The majority of positive detections (>80 %) were from surface-water samples, however, a larger proportion of total subsurface samples (~14 %; average depth = 3.4 m, min = 2 m, max = 5 m) resulted in positive detections as compared to surface-water samples (~5 %; Table 1). One of the positive detections was a cooler blank from a collection at Escanaba, Michigan in May 2013. Although three other samples tested positive from the May 2013 sampling event at Escanaba, none of these additional positive samples were associated with the contaminated cooler blank (i.e. the three positive samples were not in the same cooler). Because no other samples associated with the contaminated cooler blank tested positive for Ruffe DNA, we did not discard any data.

Table 1 Location, sampling effort, sample date, and number of positive detections for Ruffe DNA
Full size table

Positive eDNA detections at the Cheboygan River and Escanaba River corroborate results from conventional fisheries surveillance efforts where at least one Ruffe has been captured at each of these locations over the same time period (Table 2). Ruffe DNA was also detected at seven additional locations where conventional sampling failed to capture live Ruffe in 2012 and 2013, although, at two of these locations, Ruffe have been captured in previous years (site 1 starting in 2006 and site 9 in 2008). Positive detections also occurred from five locations for which conventional surveillance for Ruffe (or with gear capable of incidentally capturing Ruffe) has not been reported. There were no instances where Ruffe were captured using conventional methods in 2012 or 2013 and we subsequently failed to detect Ruffe eDNA.

Table 2 Comparison of eDNA versus conventional fisheries surveillance (from both dedicated and incidental capture efforts) for all sites where eDNA surveillance was conducted in 2012 and 2013
Full size table

Of the 72 samples that tested positive for Ruffe DNA, 40 tested positive in the initial assay with three technical qPCR replicates. The remaining 32 (including the positive cooler blank) detected Ruffe DNA only after a secondary screening of six additional technical qPCR replicates. For sites where Ruffe DNA was detected, detection failure (i.e. “initial false negative”; the percent of samples that tested positive for Ruffe DNA only after a secondary screening of additional DNA extract) ranged from 0 to 57 % (Table 3). On average, approximately 9 % of all the samples taken from locations where Ruffe DNA was ultimately detected failed to detect Ruffe DNA during the initial screening.

Table 3 Total number of samples, number testing positive for Ruffe DNA, and percent positive samples for both the initial screening (i.e. with three technical qPCR replicates) and the re-screening (i.e. with six technical qPCR replicates) for eDNA samples collected in 2013
Full size table

Discussion

The positive eDNA detections reported here are consistent with the pattern of natural spread of Ruffe, including their ongoing advance towards important ballast-water uptake areas like the St. Marys locks in eastern Lake Superior. In accordance with predictions from a ballast-mediated dispersal model (i.e. Sieracki et al. 2014) our results also provide the first indication that Ruffe is present in southern Lake Michigan, which suggests that Ruffe could be much closer to the Mississippi River than has been indicated by surveillance with conventional sampling gears alone. The potential consequences of a widespread invasion of Ruffe in the lower Great Lakes and the Mississippi River basin (see below) suggest that these results warrant consideration of a management response. However, the uncertainty associated with interpretation of positive eDNA detections (sensu Darling 2014) may impede effective decision-making and, given the reasonable fear of wasting resources if results do not indicate the presence of fish, resource managers may be reluctant to initiate expensive response efforts based only on positive eDNA detections (Finnoff et al. 2007; Darling and Mahon 2011). To allay management concerns, sources of error and uncertainty and the strength of evidence for the presence of live fish need to be communicated (Darling 2014). We attempt to put the results reported above into that context. We present our eDNA surveillance results within a conceptual two-dimensional management response framework based on weight of evidence for species presence and species spread potential (Fig. 2).

Fig. 2
figure 2

A conceptual management framework for responding to eDNA surveillance results. Response actions can be influenced by both the weight of evidence in favor of species presence and the likelihood of secondary spread. The strongest eDNA evidence in favor of species presence is evidence that, agrees with predictions from natural spread and/or dispersal models (i.e. presence of an invasion pathway), indicates a pattern of repeated detection over time, and/or corroborates any record(s) of historical live capture. The potential for spread could include consideration of presence of vectors for secondary spread, proximity to vectors, and potential for natural dispersal

Full size image

Our conceptual model is based, in part, on the idea from Jerde et al. (2011) that a gradient of evidence for species presence exists that is related to number and frequency of eDNA observations. We make the assumption that evidence to support species presence is stronger when a pathway of invasion (including natural dispersal) exists and, thus, we incorporate invasion potential as part of the weight of evidence axis. The second axis considers potential for spread from the detection site (e.g. presence of vectors for secondary spread, proximity to vectors, and potential for natural dispersal). Though not explicitly considered in our conceptual model, the potential for negative impacts of species establishment at a site could also be considered as an additional axis, including measures of site vulnerability (i.e. ecological or economic values that could be impacted if the site is invaded, sensu Margules and Pressey 2000).

In three of the six locations where we detected eDNA from more than one sample, our results simply confirm previous records of live fish from the site (sites 1, 8, 24). At Pendills Creek and Waiska River, multiple positive detections are consistent with continued eastward expansion of Ruffe along the southern shoreline of Lake Superior and also probably indicate the presence of live fish (sites 4 and 5; Fig. 3). Pendills Creek and Waiska River are both within 50 km of the Tahquamenon River and, thus, well within range for natural dispersal of Ruffe based on a conservative estimate of Ruffe dispersal distance (~25 km/year; Sieracki et al. 2014).

Fig. 3
figure 3

The conceptual management framework as applied to the eDNA survey results from 2012/2013 Ruffe eDNA surveillance effort (circled numbers correspond to site numbers listed in Table 1). Ruffe dispersal to all sampling locations was conceivable based on natural spread or dispersal model predictions

Full size image

Multiple positive detections of Ruffe DNA in southern Lake Michigan at Calumet Harbor suggest a major range expansion, but this result is consistent with predictions from a ballast-mediated dispersal model that indicates high probability of Ruffe introduction in southern Lake Michigan ports (Fig. 3, site 17; Sieracki et al. 2014). We cannot rule out the possibility that the eDNA we detected in Calumet Harbor was introduced into the system via discharged ballast water sourced from a Ruffe-invaded port, but that possibility seems unlikely for two reasons. First, eDNA degradation studies using ambient water suggest that the majority of suspended Ruffe DNA would degrade in the time required to transit from the nearest Ruffe invaded port (~24 h; Barnes et al. 2014). Second, although seven commercial vessels entered the port at Calumet in the 1-month period prior to our survey in Calumet Harbor (on July 8, 2013), none of these vessels was traveling directly from a port where Ruffe have been reported (National Ballast Information Clearinghouse 2014). Thus, it seems likely that the DNA detected in Calumet indicates the presence of live Ruffe in the harbor. Any future eDNA sampling that produces repeated detections over time, especially during intervals with no ballast-water discharge, would strengthen this conclusion (Fig. 2). A large number of eDNA samples collected from Calumet Harbor over a 2- to 4-week window could be especially informative, as such sampling would provide the spatial and temporal replicates required to resolve whether positive detections are a result of a pulse, non-fish vector or indicative of a sustained source of live fish.

At a number of sites, Ruffe DNA was recorded from a single sample, thus, representing the weakest evidence for the presence of live fish (Fig. 2). We cannot rule out that some of these detections may have arisen from low levels of contamination, as can be expected occasionally even when observing strict quality assurance protocols (Apfalter et al. 2005; Turner 2015) and as evidenced by the detection of Ruffe DNA in one of our cooler blanks. However, we found no evidence of systemic contamination in our laboratory or field control samples and all sites where a single positive sample was detected are locations where Ruffe have been captured previously (i.e. site 9) or are within approximately 100 km of sites where live fish have previously been collected and, thus, within range for natural dispersal of Ruffe (e.g. sites 20–22 are plausible given their proximity to the Escanaba and Menominee rivers where Ruffe were first captured in 2002 and 2007, respectively). Together, these observations suggest contamination had little effect on the patterns of detection reported here. It is also a possibility that some of these positive detections are a result of discharged ballast water sourced from a Ruffe-invaded port, although in most cases this seems unlikely given that most of the positive detections we report occurred at sites without an active port or in habitats upstream of existing port facilities. One exception is the positive detection at Fox River (WI), where the sample was taken near the mouth of the river and adjacent to the shipping channel.

Darling and Mahon (2011) make the case that managers should be wary of the risks associated with a failure to detect DNA of target AIS when it is present (i.e. false negative), given that highly sensitive eDNA surveillance methods make detection of incipient invasions a real possibility and, thus, increase the potential to minimize long-term control costs and impacts if establishment can be prevented. In this study, we screened samples using a quantitative PCR assay to reduce the rate of false negatives (i.e. relative to less sensitive traditional PCR methods; sensu Thomsen et al. 2012a, b; Wilcox et al. 2013; Nathan et al. 2014; Turner et al. 2014a, b). Yet, after doubling our level of replication we detected numerous additional positive samples (Table 3). In the Tahquamenon River, where Ruffe have been collected since 2006, 12 out of the 29 samples that eventually tested positive for Ruffe were negative during the initial qPCR screening (3 replicates), suggesting that the concentration of target species DNA is low and highly patchy both within an extracted sample and in the environment (i.e. 25 % of samples collected tested negative despite Ruffe presumably being established at low densities at this site). Patchiness of eDNA is consistent with the occurrence of a substantial portion of eDNA in relatively large particles (i.e. particle sizes corresponding to cells or clumps of cells; Turner et al. 2014a, b). Furthermore, many locations had a relatively low number of positive sample detections (<5 % of samples positive at a location). This is expected if there are few Ruffe releasing eDNA and imperfect mixing of the water column (Jerde and Mahon 2015). If we assume that Ruffe are physically present in these locations with very few positive detections, then presumably there is a large false negative rate (see “Appendix 2: A method for estimating the sensitivity of the Ruffe eDNA survey” section). This is a critical consideration in locations where no positive detections were recorded and where there was a limited sampling effort (locations with fewer than 30 samples collected). The results from our re-screening are consistent with other eDNA studies that highlight the importance of PCR replication and the potential for high prevalence of false negatives from eDNA samples when detection probabilities are low (Ficetola et al. 2015; Furlan et al. 2015).

It follows that decision makers may be more willing to initiate potentially expensive and long-term response programs, where the cost of the failure to act is high (Darling and Mahon 2011). Although effects of Ruffe populations in Lake Superior are equivocal, it is possible that impacts on food webs in other parts of the Great Lakes or connected basins could be larger, especially if populations become large or widespread (Bronte et al. 1998; Gunderson et al. 1998). Ruffe can compete with native Great Lakes fishes for space and food (Edsall et al. 1993; Sierszen et al. 1996; Ogle et al. 2004; Fullerton and Lamberti 2005). It has also been suggested that complete colonization of the Great Lakes by Ruffe could result in a decrease in Yellow Perch populations by 10–60 % and of Whitefish and Walleye populations by up to 25 % with accompanying annual losses in the range of $24 to $214 million in sportfishing and commercial fishing revenue (Leigh 1998). In addition, the establishment of Ruffe in the southern basin of Lake Michigan would likely accelerate spread out of the Great Lakes and into the Mississippi River basin via the Chicago Area Waterway System. Ruffe was identified as one of 29 species (and one of six fishes) with the potential to cause moderate to severe effects if introduced into the Mississippi River basin (Jerde et al. 2010; US Army Corps of Engineers 2014). The extent of the threat to the Mississippi basin is as yet unquantified, but globally significant freshwater biodiversity is potentially at risk because the Mississippi River and its tributaries contain the largest number of freshwater fishes (260 species) of any region at comparable latitudes (Smith 1981; Fremling et al. 1989) and are a global center of diversity and endemism for crayfish (Lodge et al. 2012) and unionid mussels (Abell et al. 2000). Some native fishes serve as important hosts for endangered unionid mussels and a subset of these host fishes rely heavily on benthic food sources (e.g. Freshwater Drum, Sauger, and Catfishes), and are expected to compete directly with Ruffe (Ed Rutherford, NOAA GLERL, personal communication, December 10, 2013).

Given the detection of Ruffe eDNA in Calumet and the potential risk to the Mississippi River basin there may be value in implementing ongoing surveillance for Ruffe with both eDNA and conventional sampling methods, and managers might consider adopting measures to contain and prevent the movement of Ruffe out of the Great Lakes via the Chicago Area Water System (Fig. 3). In addition, the detection of Ruffe eDNA near important ballast-water uptake areas (sites 7 and 21; Jennifer Sieracki, National Park Service, personal communication, June 24, 2015) highlights further the invasion risk posed by the movement of ballast water within the Great Lakes (Keller et al. 2011; Adebayo et al. 2014; Sieracki et al. 2014). Thus, other slow-the-spread and control options proposed early in the Ruffe invasion, including ballast- water management, might also be productively revisited by managers (Busiahn 1997).

While the capture of a live specimen will always be more compelling than eDNA evidence, we caution against dismissal of eDNA results in the absence of corroborating live captures, especially in the absence of an explicit examination of relative sampling effort and detection sensitivity of conventional gears. There is increasing evidence that genetic surveillance methods are more sensitive than conventional approaches (Jerde et al. 2011; Thomsen et al. 2012a, b; Turner et al. 2012; Biggs et al. 2015; Valentini et al. 2015), which means that eDNA detections are possible even when sampling with conventional gear fails—especially at the putative invasion front when target organisms are rare (Darling and Mahon 2011; Jerde et al. 2011; Dejean et al. 2012). Controlled experiments that directly compare the sensitivity of eDNA sampling and conventional sampling methods when target species are at low abundance would be helpful (Darling and Mahon 2011). Fuller consideration of dispersal, invasion pathways, and other factors affecting invasion risk would provide a stronger basis for management response decisions following the eDNA detection of a potential incipient invasion. Only when eDNA detections are considered alongside other lines of evidence (including probability of spread and potential for negative impacts) can they help guide appropriate and defensible management decisions.