Invasive species can have negative effects on native biodiversity and ecosystem function, and suppression is often required to minimize the effects. However, management actions to suppress invasive species may cause negative, unintended effects on non-target taxa. Across the USA, lake trout (Salvelinus namaycush) are invasive in many freshwater ecosystems, reducing native fish abundance and diversity through predation and competition. In an integrated pest management approach, lake trout embryos in Yellowstone Lake, Wyoming are suppressed by depositing lake trout carcasses onto spawning sites; the carcasses reduce dissolved oxygen concentrations as they decay, causing embryo mortality. We conducted a field experiment during one ice-free season at four sites in Yellowstone Lake to investigate the non-target effects of carcass treatment on benthic invertebrates, which could have consequences for native fish diets. While overall invertebrate density and biomass did not respond to carcass treatment, Chironomidae midges and Sphaeriidae fingernail clams decreased in abundance. Carcass treatment altered invertebrate community structure based on density, but not biomass. Carcass treatment to suppress invasive fish embryos has spatially localized, non-target effects on some benthic invertebrate taxa. Given the small spatial extent of carcass treatment within the lake, we conclude it is unlikely that carcass treatment will alter food availability for native fishes.

Study area

Yellowstone Lake is located in Yellowstone National Park in northwestern Wyoming at an elevation of 2,357 m (Fig. 1). With a surface area of 340 km², it is the largest high-elevation lake in North America and is generally ice-covered from late December to mid-May. Yellowstone Lake is a mesotrophic, dimictic lake that thermally stratifies in the summer (Kilham et al. 1996). It has complex bathymetry, with a mean depth of 43 m, a maximum depth of 148 m, and hydrothermal vents distributed throughout the northern and western regions of the lake (Kaplinski 1991). Benthic invertebrate assemblages are dominated by two amphipod genera, Hyallela and Gammarus, which together comprise approximately 55% of invertebrate biomass (Wilmot et al. 2016).

Lake trout are a large, piscivorous species that is native throughout Canada and parts of the northern United States. Lake trout were first discovered in Yellowstone Lake in 1994 and reached peak abundance in 2012, with an estimated adult population of nearly 1 million individuals (Koel et al. 2020a). The lake trout suppression program in Yellowstone Lake began in 1995 and currently removes about 300,000 adult lake trout each year (Koel et al. 2020a). More than 4 million were killed between 1995 and 2021 with a majority of the carcasses deposited into deep areas (> 70 m) of the lake (Koel et al. 2022). This extensive suppression effort has reduced adult lake trout abundance by > 80% since 2012, however, recruitment of young lake trout remains high (Koel et al. 2020a). The lake trout spawn from late September to early October (Heredia et al. 2021) at sites characterized primarily by cobble and bedrock substrate; confirmed lake trout spawning sites range in size from 0.3–2.0 ha and comprise 0.03% of the total area of Yellowstone Lake (Koel et al. 2020b). To curtail lake trout recruitment and improve suppression efficiency, the NPS has developed a novel method that deposits carcasses directly on spawning areas in the littoral zone of the lake to induce decomposition and cause mortality in lake trout embryos via oxygen depletion (Thomas et al. 2019; Poole et al. 2020).

Carcass treatment

Carcass treatment occurred at two lake trout spawning sites, Flat Mountain Hump and Snipe Point, from 12 August to 2 October 2019 (Fig. 1). Of the 14 confirmed lake trout spawning sites on Yellowstone Lake (Koel et al. 2020b), these two sites were selected because were deep enough and far enough from shore to avoid attracting terrestrial wildlife (Thomas et al. 2019) and were adjacent to frequently targeted suppression netting locations to facilitate the transportation of lake trout carcasses. Carcass treatment only occurred at two spawning sites due to limited availability of lake trout carcasses during the autumn and logistical difficulties of transporting carcasses across a large lake. Two control sites, Flat Mountain Elbow and Elk Point, were selected based on similar depths, substrate type, and logistical considerations, including avoiding sites frequently targeted by suppression netting. Substrate at carcass treatment and control sites was dominated by bedrock or cobble (< 250 mm), and the sites varied from 2.5 to 9.5 m deep. Limited carcass treatment occurred at both treatment sites in 2018 as part of a pilot study, but carcass coverage was extremely low (< 3% coverage in October 2018). When sampling began in June 2019, no carcass material was observed at Snipe Point, and small amounts were observed at Flat Mountain Hump (< 1% coverage). Carcass treatment sites were marked with buoys anchored to the substrate with concrete blocks. Gill netting crews dumped whole and shredded lake trout carcass material from boats within 5 m of marker buoys. Carcass dumps occurred opportunistically when gill netting crews had been fishing near the carcass treatment sites. Approximately 6,000 kg of fish carcass material were deposited at each carcass treatment site.

To measure DO, scuba divers secured one miniDOT logger (Precision Measurement Engineering) to the substrate surface at each carcass treatment and control site. The loggers recorded DO (mg/L) at 60-minute increments throughout the entire sampling period (17 June–1 October 2019). See Supplement A for details on methods and data analysis.

To monitor carcass cover at the carcass treatment sites, we photographed a 1 m² quadrat placed on the substrate surface using a GoPro underwater camera. We took five adjacent photographs of the quadrat along a 0º heading, starting directly north of the concrete anchor marking the site, and we took five additional photographs along a 180º heading, directly south of the concrete anchor. We used ImageJ (Rashband 2018) to calculate the percent area of each quadrat covered by carcass material and averaged across all 10 quadrats to estimate mean carcass cover every two weeks during the carcass treatment period, resulting in four measurements per site.

Invertebrate collection

At each site, triplicate benthic invertebrate samples were collected monthly for three months before carcass treatment started (17 June–11 August 2019) and every two weeks (4 sampling intervals) after carcass treatment started (12 August–1 October 2019). To quantitatively sample benthic invertebrates, we used a Scuba diver-operated suction sampler constructed from an electric bilge pump mounted on a plastic cutting board with a 500-µm mesh collection net (Cross et al. 2011). We randomly placed a 0.25 m² quadrat on the substrate surface and used the suction sampler to collect invertebrates within the quadrat. While we did not use a quantitative method to randomly select quadrat placement locations, we placed the quadrat where the substrate was undisturbed by divers, and the same diver placed the quadrat for each sample to reduce sampling variability. After surfacing, we rinsed all contents of the collection nets into a 500-µm sieve and preserved all material retained by the sieve in 75% ethanol.

In the laboratory, we subsampled invertebrate samples to the smallest fraction that included ≥100 individuals using a plankton splitter. Subsample fractions ranged from ½ to 1/16. We sorted invertebrates from debris and organic material and identified individuals to the lowest practical taxonomic level, which was often genus or family and occasionally order (Rabeni and Wang 2001; Merritt et al. 2008). We measured 25 randomly selected individuals of each taxon and used published length-mass regressions to calculate biomass (AFDM) of each taxon (Bottrell et al. 1976; Benke et al. 1999; Méthot et al. 2012).

Data analysis

To determine if invertebrate density and biomass changed in response to carcass treatment, we assessed the interaction between time (a categorical variable with 2 levels: before and during treatment) and treatment using linear mixed-effects or zero-inflated negative binomial (ZINB) models. Our models included treatment, time, and their interaction as fixed effects, and sampling date nested within site as random effects to account for temporal and spatial autocorrelation. We used linear mixed effects models for most metrics, and we natural log-transformed response variables when necessary to meet model assumptions. Because we expected the control and treatment sites to change over time due to seasonal variation, we used the interaction between time and treatment as evidence of a treatment effect. We focused on the three most abundant taxa (Gammarus sp. and Hyallela sp. amphipods and non-Tanypodinae Chironomidae; Table S1), hypoxia-sensitive taxa (Ephemeroptera and Trichoptera), and sessile taxa (Sphaeriidae). We chose Ephemeroptera and Trichoptera to represent hypoxia-sensitive taxa instead of a metric including Ephemeroptera, Trichoptera, and Plecoptera (EPT, Karr 1991) because Plecoptera are not present in Yellowstone Lake. Because some taxa were present at control sites but not carcass sites, our analyses of Ephemeroptera and Trichoptera density and biomass combined taxa within these two orders and only included taxa observed at control and carcass sites. We used ZINB models for Ephemeroptera and Trichoptera density and Sphaeriidae density, because these data were overdispersed and had many zeros (Brooks et al. 2017). We used Akaike information criterion corrected for small sample sizes (AICc) to select ZINB models over other possible models (Table S2). We did not perform statistical analysis on hypoxia-sensitive taxa (i.e., Ephemeroptera and Trichoptera) biomass or Sphaeriidae biomass because ZINB models are only appropriate for count data (such as abundance), and these overdispersed data did not meet model assumptions for other types of analytical tools. Statistical significance was tested with an alpha value of 0.1 because of variation expected from such a large-scale experiment with low replication, to improve our detection of invertebrate responses, and to reduce the possibility of Type II error. Model assumptions were examined and met. All statistical analyses were conducted in R version 3.6.2 (R Core Team 2019) using the lme4 package for linear mixed effects models (Bates et al. 2015) and the glmmTMB package for zero-inflated negative binomial models (Brooks et al. 2017). 

We used non-metric multidimensional scaling (NMDS) based on Bray-Curtis dissimilarities to visualize how invertebrate community structure based on untransformed density and biomass responded to carcass treatment (Kruskal 1964; Minchin 1987). To test if invertebrate community structure changed in response to carcass treatment, we used a two-way mixed effects permutational multivariate analysis of variance (PERMANOVA) including time, treatment, and their interaction as fixed effects and site as a random effect (Anderson 2014). We used similarity percentages (SIMPER) analysis to identify the taxa that contributed most to differences in multivariate position by group, which allowed us to interpret changes in community structure (Clarke 1993). Community analysis was conducted using the vegan package (Oksanen et al. 2019).

Microsoft Excel