Goldfish (Carassius auratus), a globally invasive species, is increasing in abundance in lakes and ponds within the Laurentian Great Lakes region. We hypothesized that selection pressures imposed on feral goldfish populations that have adapted to urban and natural ponds would result in elevated ecological performance compared to cultivated pet-store goldfish. We tested the hypothesis by comparing the functional response (foraging efficiency) and temperature tolerance (critical thermal maximum, CTmax) of feral and cultivated goldfish under current (18°C) and projected future (26°C) surface water temperatures. After acclimation to projected climate conditions, all goldfish populations displayed higher maximum feeding rates and temperature tolerances. Feral goldfish had higher maximum feeding rates than cultivated goldfish at both temperature treatments, indicative of higher trophic impact potential. In contrast, the CTmax of goldfish did not vary significantly among populations. Collectively, these results indicate that temperature and intraspecific variation influence the foraging efficiency of goldfish and thus could mediate their ecological impacts. Population-level variation in foraging suggests that selection in urban and natural ponds may increase the ecological impact of introduced goldfish.

Fish sampling, measurements, and husbandry

Goldfish were collected from three urban stormwater management ponds and two natural waterbodies between May 17 and August 15, 2024 (Table S1). Fish were collected using seine nets (<6 mm mesh), backpack electrofishing, and boat electrofishing. Cultivated goldfish were purchased from a pet store in Montreal, Quebec (Centre d’Animaux Nature) in two shipments. Thirty fish were collected from each pond/waterbody (150 total), and 60 fish were purchased from the pet store. Fish were size-matched using standard length to account for size-dependent variation in feeding rate (Rall et al., 2012). All individuals collected were 4.7–8.2 cm in standard length (SL). Fish of these sizes have been found to be between 3–12 months of age in previous studies of goldfish growth rates (Mugiya et al., 1981; Lorenzoni et al., 2007).

Fish were transported in aerated coolers from sampling sites to McGill University, where they were housed in 91-Litre aquariums equipped with filters, air stones, gravel substrate, and artificial vegetation in temperature-controlled chambers set to 22°C. All care procedures followed standard animal-care protocols (CCAC SOP 519 and McGill University AUP 8267). Five fish were housed per tank to allow social interaction at required stocking densities. Water-quality parameters were tested weekly to ensure ranges remained within acceptable ranges, with 30% water changes being performed each week or when indicated by high parameters (ammonia, nitrite, nitrate). Fish were fed the equivalent of 0.5% of their body weight daily using Hikari Sinking Goldfish Pellets. A daily feed of 0.5% of fish body weight has been determined to be the most appropriate density for goldfish husbandry (Trueman 2021).

Experimental design

Fish were acclimated to laboratory conditions for one week before temperature changes began. The housing temperature for half of the fish from each population was reduced to 18°C (current mean summer temperature for nearshore areas of the lower Great Lakes basin); the other half was increased to 26°C (predicted mean maximum summer temperature under climate change; Trumpickas et al., 2009, 2015), at a rate of 1°C per day. One week after target temperatures were reached, fish were individually marked to distinguish them from tankmates (n=5) by clipping the top and/or bottom corner of the caudal fin at a 45-degree or straight angle. A pilot experiment found no significant difference in foraging performance between clipped and unclipped fish. Fish were left to acclimate to temperature conditions for 2 weeks before experiments began. The experimental design included 2 experiments, a functional response experiment and a critical thermal maximum (CTmax) experiment (Figure 1). Fish used for the CTmax experiment were reused from the functional response experiment after a 2-week rest period.

Comparative functional response (FR) experiments

The functional response of goldfish from two cultivated shipments and five feral populations were tested using fish acclimated to two temperature regimes: 18°C and 26°C. This resulted in 14 experimental groups (7 populations/shipments × 2 temperature treatments). Fifteen fish from each population were tested for each temperature treatment, with the exception of Cootes Paradise Marsh, where 10 fish were tested per temperature treatment due to mortalities.

Individual fish were removed from holding aquaria and mass (g) and standard-length (cm) measurements were taken. Fish were transferred to 45 L experimental tanks, where they were left to acclimate for 24 hours (Avlijaš et al. 2022). Plates with glued ≈5 cm coarse gravel covered the tank bottom during functional response trials to provide complexity and interstitial refuge for prey (South et al. 2019). Fish were starved for 24 hours to ensure homogeneity in hunger levels across trials. A common food item, chironomid larvae (thawed Chironomus bloodworms from frozen stock), were used as prey, with n=5 replicates at n=8 densities (3, 6, 9, 18, 36, 72, 144, 288). Multiple small prey densities (3–9 bloodworms) were chosen to allow for the possibility of a prey refuge, i.e., a Type III functional response curve.

Forty trials were performed per experimental group (5 replicates × 8 densities), resulting in 560 trials in total (14 experimental groups × 40 trials). The prey density tested for each fish was randomized. In the event of fish re-use, individuals were rested for one week and tested at a different prey density (Reid and Ricciardi 2022). Individual fish were retested 2–3 times (4 times for fish from Cootes Paradise Marsh due to mortalities). Frozen chironomid larvae were thawed, counted, and distributed evenly across the aquarium bottom when trials began. Fish were left alone to feed for one hour (Guo et al. 2017). Upon trial completion, they were promptly removed from experimental tanks and returned to holding tanks to allow collection and count of the remaining prey.

Functional response data were analyzed in R (R version 4.4.2, RStudio version 2024.09.1) using the Frair package (Pritchard et al. 2017) alongside the ‘glmmTMB’ package. The type of functional response curve (I, II, or III) was determined using 1) the Frair_test function, 2) Akaike information criteria (AIC), and 3) visual inspection. The Frair_test function fits a polynomial equation to the proportion of prey consumed as seen in Juliano (2001) and selects the curve based on the direction (+/-) of the first-order term (density). A negative first-order term indicates a Type II functional response curve, whereas a positive first-order term indicates a Type III. AIC was compared by fitting models with a scaling exponent (q), parametrized as q=0 equals Type II functional response, resulting in a null hypothesis test. AIC values of the flex model (flexpnr), Type II model (rogersII), and Type III model (hassIIInr) were compared, with the model with the smallest AIC value being preferred. Preference for a flex model in which evidence q is higher from 0 is indicative of Type III functional response. Models were also checked using visual inspection of a graph plotting initial prey density (x-axis) against the proportion of prey consumed (y -axis). A decreasing graph line indicates Type II, and an increasing one indicates Type III.

After the model type was determined, attack rate (a) and handling time (h), with standard errors, were extrapolated from the chosen frair_fit model. The maximum feeding rate was calculated as 1/hT, where T is time (1 hour), and the functional response ratio (FRR) was calculated as a/h. Non-parametric bootstraps (n=999) were generated using the frair_boot function to provide 95% confidence intervals around FR curves. These curves were assessed visually for overlap. As data were fitted to the same curve, the frair_compare function was used to calculate differences in attack rates and handling times. Type III sum of squares (ANOVA) of a generalized mixed effect model (GLMM) was used to test the impact of source (wild, urban, pet store), temperature and the source × temperature interaction, and fish length on the number of bloodworms eaten at the highest prey density (n=288). Population, experimental tank, and fin clip were used as random effects. Differences in fish body length between sources was determined with a one-way ANOVA test.

Comparative critical thermal maximum (CTmax) experiments

The upper thermal tolerance of goldfish was tested using fish reused from functional response experiments (Cootes Paradise Marsh, Stormwater Pond 90, London Stormwater Pond, Cootes Paradise Marsh, pet store). The fish rested for two weeks after their last functional response trial, resulting in a total acclimation period to 18°C or 26°C of 1.5 months. The critical thermal maximum was tested following an established protocol (McDonnell and Chapman 2015; Wells et al. 2016). Fish were placed in clear baskets in 23-Litre tanks with a Julabo heater circulator. Fish were allowed to habituate to experimental conditions for 1.5 hours. During trials, the temperature was increased at a rate of 0.3°C per minute (Becker and Genoway, 1972). The CTmax was recorded as the temperature at which fish lost equilibrium for 3 seconds (Beitinger et al. 2000). CTmax data were analysed in R (R version 4.4.2, RStudio version 2024.09.1) using the ‘cars’ and ‘lme4’ packages. Type III sum of squares (ANOVA) of a generalized linear mixed model (GLMM) was used to determine the impact of population and temperature on critical thermal maximum. The Gamma family with a log link function was used to accommodate non-continuous responses and repeated observations. Trial number, experimental basket used, fish length, and fin clip were included in the model as random effects. The acclimation response ratio (ARR), a measure of how thermal acclimation affects the upper tolerance limits of an animal, was calculated as follows: ΔCtmax / ΔTa, where Ta is the acclimation temperature and all units are in degrees Celsius. A high ARR represents a greater change in response dependent on acclimation temperature.