Use of environmental DNA (eDNA) in streams to detect feral swine (Sus scrofa)

Study sites

This study was conducted in two free-flowing streams from August 2017 to April 2018. Bluff Creek is located near Sanford, Michigan (16N 01W Section 34, 43.72897 N, −84.47174 W) and Black Creek near Alamando, Michigan (16N 01W Section 31, 43.73348 N, −84.42216 W), both in Edenville Township, Midland County (Fig. 1). The Michigan Department of Natural Resources (MDNR) issued a state land use permit for access (#73-2017-013). Cover types included lowland shrubs, marshes, wetlands, beaver (Castor canadensis) ponds, and swamp hardwoods interspersed with upland oak (Quercus spp.) and aspen (Populus spp.) stands. Bluff Creek is characterized as uniform and free flowing with little streambank vegetation undergrowth, a full over story tree canopy, and narrow floodplain, with steep sloping banks that parallel the stream. Black Creek has a wide floodplain with multiple water channels when stream flow is high, and thick understory brush along the streambank.

Monitoring for feral swine

Because the study design required artificially introducing swine DNA into streams at specific locations and times, inferring absence of feral swine within the study area was essential. Previous research in Michigan monitored the study area since 2013, with no evidence of feral swine detected (D Etter, pers. comm., 2017). The MDNR logs all reports of feral swine in the state to Township, Range and Section, and no reports were recorded for feral swine in Edenville Township from 2016 to 2018 (D Etter, pers. comm., 2019). Additionally, the Michigan Pork Producers had no records of any commercial swine operations within the immediate vicinity of the study sites (M Kelpinski, pers. comm., 2019). From November 2016 to April 2018, one week prior to DNA insertion and water sampling, field surveys were conducted along each stream for feral swine sign (tracks, scat, wallows, rooting; Mayer, 2009). Additionally, trail cameras (Bushnell® Trophy Cam HD Aggressor 14MP Trail Camera; Bushnell Outdoor Products, Overland Park, Kansas) were used to monitor wildlife trails and beaver dams along Bluff and Black Creeks. Ten cameras spaced 400 m apart were set over a total distance of 4000 m on each creek. Five cameras were placed upstream and five downstream from the DNA source point. Ten additional cameras were set in the Mud Creek watershed, located between Bluff and Black Creeks, for additional surveillance. Trail cameras were checked once every two weeks for the entirety of the study and reviewed for photos of feral swine.

Field sampling methods and materials

The midpoint of both stream sampling transects was strategically selected for road or trail accessibility by truck or ATV. The sampling scheme was then based on equal distribution of trail cameras up and downstream from the pre-determined midpoint (200 m; Fig. 2). If a road or bridge occurred at the midpoint, the location was moved upstream so the end of the sampling transect (400 m) ended ≥100 m upstream from the man-made structure.

Metal fence posts were placed on opposite streambanks to mark the location for artificial insertion of swine DNA. Additionally, one post was placed every 100 m downstream for a total of 400 m to mark the start location of each 100 m segment. Flagging was used along the stream bank to mark 10 m increments within each segment and these locations were GPS recorded (Garmin GPS 12 XL; Garmin Ltd., Olathe, Kansas; Fig. 2).

There is a long history of interbreeding among domestic and wild pigs (McCann et al., 2018), therefore, feral and domestic swine DNA could not be differentiated on the molecular level at the time of this study. Therefore, DNA specimens (legs and/or hides) were acquired from feral or domestic swine culled by the United States Department of Agriculture—Wildlife Services in Michigan, or from a swine processing facility. Each specimen was frozen and subsequently thawed for 24 h before insertion into a stream. To simulate a swine crossing a stream, specimens were attached to a rope pulley system between the fence posts and pulled through the stream four times at the source point, and then the specimen was bagged, removed from the site, and disposed.

Water samples were collected with at least 18 days between each DNA insertion, ensuring DNA degradation in the environment between sampling events (Dejean et al., 2011; Thomsen et al., 2012; Piaggio et al., 2014). Samples were collected from the water surface in the middle of the stream using a telescopic dipper (1.5–3.5 m Conbar Telescopic Dipper; Forestry Suppliers Inc., Jackson, Mississippi). The ladle was dipped in an upstream motion while collecting the minimum amount of water needed (∼50 ml), and extra stream water was placed in an 18.9 L bucket for disposal. To avoid cross contamination of samples between subsequent sampling of each 100 m segment, the telescopic dipper was sterilized with a 20% bleach solution (Goldberg et al., 2013; Jane et al., 2015; Williams, Huyvaert & Piaggio, 2016) for 2–5 min then rinsed multiple times with distilled water into the 18.9 L bucket.

Sterivex filters (Sterivex—GP capsule filter pore size 0.22 µm; Millipore Sigma, Darmstadt, Germany) were used for collecting and transporting DNA based on prior eDNA extraction efficiency studies (Spens et al., 2017). Because filters can clog when sampling turbid waters, 2 Sterivex filters were used for collection (250 ml per filter; Piaggio et al., 2014). To ensure that swine DNA was not naturally occurring, prior to swine DNA being manually introduced into streams 500 ml of water was strained through filters at the source point (designated as T = 0). To determine whether DNA was detectable at the source point immediately after DNA insertion, another 500 ml of water was strained through filters (designated as T = 1). This process was repeated two hours later (T = 2) to assess if DNA persisted at the source point. After T = 2 sampling, collection continued downstream for a total of 400 m (Fig. 2). Every 10 m within a 100 m segment, one 50 ml sample was strained through a filter for a total of 500 ml collected from each 100 m segment. Sterivex filter water samples were filled with Longmire’s lysis buffer using a 3 ml syringe (Renshaw et al., 2015) to preserve DNA captured in the filter, capped, labeled individually, and then wrapped with parafilm. Filters were stored at room temperature until DNA extractions could be completed. After each sampling event, all samples were directly transported to the genetics laboratory for molecular processing.

Stream measurements

Water temperature, velocity, pH, and turbidity were measured during each DNA sampling event. Water temperature, pH, and turbidity were collected from shore using the telescopic pole and dipper during collection of water samples. Temperature and pH meters (Oakton pH 5+ and 6+ Meters, pH/Temperature/mV; Oakton Instruments, Vernon Hills, Illinois) were calibrated prior to sampling each day and data were recorded at the source point and at every 100 m segment downstream (Fig. 2). Turbidity, measured in nephelometric turbidity units (NTU), was monitored using a turbidity meter (Oakton Turbidity Meter; Oakton Instruments, Vernon Hills, Illinois) by collecting a small water sample (>10 ml) in a small glass vial and placing it into the meter. The turbidity meter was calibrated before each sampling event. Turbidity was collected from the middle of the stream at the source point and every 100 m downstream (Fig. 2). The stream bottom or water flow could not be disturbed by researchers during DNA sampling so velocity was collected at the end of each sampling event. Velocity was measured using a flow probe and average velocity function (Global FP111 Flow Probe; Geo Scientific Ltd., Vancouver, British Columbia, Canada). Average velocity was determined by moving the flow probe smoothly and evenly back and forth from top to bottom of the water column for 20–40 s. Sampling the 400 m transect took an average of 4 h 45 min and changes may have occurred in flow from the beginning of a sampling event to the end.

Laboratory sample processing

DNA extraction and digital droplet polymerase chain reaction (ddPCR) (BioRad QX200 Droplet Digital PCR System; BioRad Laboratories Inc., Hercules, California) were used to process field samples in the molecular lab. Total DNA extraction from Sterivex filters followed the protocol developed by Spens et al. (2017). Extracting DNA from the Sterivex filters necessitated approximately 48 h during which DNA was extracted from the Longmire’s lysis buffer and the enclosed filter using a Qiagen DNEasy kit (Qiagen, Inc., Hilden, Germany). During the second day of the extraction protocol, the DNA found in both the Longmire’s lysis buffer and enclosed filter were combined and held at −20 °C until they were analyzed by ddPCR. Sterivex filters underwent DNA extraction individually and then samples from the 2 filters used for collection were pooled for each individual sample event (i.e., T = 0, T = 1, T = 2, and each 100 m segment). A subsample of the pooled DNA per segment was used for ddPCR and the remaining DNA was archived at −80 °C.

The ddPCR platform partitions each sample into ∼10,000–20,000 individual replicates (droplet) for each DNA extraction analyzed. The platform targets and results in an actual count of target molecules (Nathan et al., 2014). Target Sus scrofa primers and a species specific hydrolysis probe for ddPCR were previously developed by Williams, Huyvaert & Piaggio (2017). Targeted swine DNA presence was measured by the amount of target DNA copies per microliter (copies/µl). Every ddPCR plate (i.e., each time a group of samples were screened/analyzed) included three positive swine controls from genomic DNA (tissue extraction DNA) and three negative controls (blanks; ddH2O). In all analyses, the positive control tested positive and the negative controls screened negative, demonstrating no issue with contamination in the processing of the DNA samples in the laboratory. Water blanks were included for DNA extractions, analyzed as normal samples, and no water blanks tested positive for swine DNA. For this study, any sample with a minimum of one positive droplet per µl indicated a positive sample for the presence of swine eDNA. Positive samples were replicated in order to demonstrate repeatability of the data. However, simply based on cost per sample and timing for this research, we did not repeat processing of all negative study samples.

Data analyses

Multicollinearity of predictor variables was assessed by computing a Variance Inflation Factor (VIF; Cohen et al., 2003) for each predictor variable from a fully parameterized general linear model (binomial family). The full model included swine DNA detection (binary response, yes or no) as the response variable and stream type (as a factor), distance downstream, stream velocity, water temperature, water pH, discharge, and turbidity as predictors. Predictors with VIF >5 were sequentially removed from the model, with highest VIF removed first. Distances downstream from the swine DNA source point included 0 m (T = 1), 100 m, 200 m, 300 m, and 400 m. Stream was included as a factor because Bluff and Black Creeks represented variable habitat types and stream characteristics. Parameters from the final generalized linear model were significant if 95% confidence intervals (CI) did not overlap 0. All analyses were completed using packages lme4, regclass, and ggplot2 in R Version 3.6.1 (R Core Team, 2019).