Carbon Dioxide-Induced Mortality of Four Species of North American Fishes

Effective management of invasive freshwater fishes often benefits from development of novel control techniques as alternatives or complements to conventional approaches. Fisheries managers currently utilize both mechanical (e.g., electrofishing and gill-netting) and chemical removal techniques (Finlayson et al. 2002; Shepard et al. 2002; Buktenica et al. 2013). Unfortunately, traditional chemical control agents such as rotenone and antimycin-A negatively affect aquatic invertebrates, larval amphibians, and other nontarget organisms (Vinson et al. 2010; Billman et al. 2011, 2012). Further, antimycin-A is no longer registered by the U.S. Environmental Protection Agency (USEPA), making it unavailable for use. The USEPA recently registered carbon dioxide (CO₂) for restricted use as a pesticide for deterrence and lethal control (USEPA Registration No. 6704-95) and researchers have demonstrated it to negatively impact a range of aquatic species (Bierbower and Cooper 2010; Nielson et al. 2012; Abbey-Lambertz et al. 2014; Hasler et al. 2016; Cupp et al. 2018; Robertson et al. 2018; Waller and Bartsch 2018; Waller et al. 2019). Carbon dioxide use as a pesticide is expanding and may eventually provide a potential alternative or complement to traditional chemical control agents. The benefits of CO₂ as a pesticide in freshwaters are numerous, and include its demonstrated toxicity to fish, larval amphibians, and aquatic invertebrates, natural occurrence in the atmosphere and aquatic environments, and availability to aquatic primary producers following application (Noatch and Suski 2012; Treanor et al. 2017; Hasler et al. 2018).

Exposure to elevated levels of CO₂ can have both sublethal and lethal effects on fishes (Gelwicks et al. 1998; Ross et al. 2001; Clingerman et al. 2007; Kates et al. 2012; Dennis et al. 2016; Tix et al. 2017; Tucker et al. 2019). Initially, contact with sublethal levels of CO₂ causes a decrease in body fluid pH, but the accumulation of bicarbonate ions in body fluids buffers this change. Sublethal exposure can also result in difficulties with respiration, blood circulation, and nervous system function, and long-term effects include diminished growth rate and reduced reproduction (Smart et al. 1979; Ishimatsu and Kita 1999; Fivelstad et al. 1999). Lethal exposure causes mortality through acidosis or a reduction in blood or hemolymph pH (Baker et al. 2009). Low blood pH negatively affects the ability of blood hemoglobin to transport oxygen and, ultimately, causes death.

While several studies have addressed the physiological effects of CO₂ on fishes and invertebrates (e.g., Ishimatsu et al. 2004; Kates et al. 2012; Hannan et al. 2016; Hasler et al. 2016; Jeffrey et al. 2017), the list of species with defined lethalities is relatively limited but growing (Treanor et al. 2017). Even though CO₂ is presently registered as a pesticide, future users of CO₂ as a fisheries management tool will benefit from knowledge of sensitivity thresholds for individual fish species (Noatch and Suski 2012). To that end, our primary goal was to assess the effects of CO₂ on mortality of three widely distributed fish species: Common Carp (CC) Cyprinus carpio, Channel Catfish (CF) Ictalurus punctatus, and Rainbow Trout (RBT) Oncorhynchus mykiss (Zambrano et al. 2006; Johnson et al. 2008; Weber and Brown 2011). Additionally, we sought to determine the effects of CO₂ on Westslope Cutthroat Trout (WCT) Oncorhynchus clarkii lewisi, which RBT are currently displacing across much of their Rocky Mountain distribution (Shepard et al. 2005). Determining whether managers could apply CO₂ to a body of water to induce mortality of invasive fishes while causing minimal mortality to native fishes would be important to future fish restoration projects. Our specific objectives were to 1) estimate the 24-h 50% lethal concentration (LC₅₀) and 95% lethal concentration (LC₉₅) for fingerlings of these four fish species (CC, CF, RBT, and WCT) and 2) identify differences among species in tolerance of CO₂ concentration.

We conducted all laboratory trials at the Bozeman Fish Technology Center (BFTC; U.S. Fish and Wildlife Service [USFWS]), Bozeman, Montana. We captured CC used in the trials in the wild, while we transported RBT, WCT, and CG from hatchery facilities. We transported CC in an oxygenated transportation tank on a BFTC vehicle. We acclimated all individuals upon arrival at the BFTC to facility spring water in fiberglass holding tanks for 24 h to the temperature range available during the trials (i.e., 10–11°C). We held CC in 76-cm-diameter tanks (228 L), CF in 91.4-cm-diameter tanks (300 L), and both RBT and WCT in 86.4-cm-diameter tanks (297.5 L). Water alkalinity was 175 mg/L CaCO₃. After this temperature acclimation, we then successfully transitioned all fish onto food and held them for 14 d prior to the start of the exposure trials. We fed all fish a commercial trout diet (Skretting USA, Tooele, UT) while in holding tanks using a timed 12-h belt feeder up until the time of trial initiation. We did not feed fish in experimental tanks during the trials.

We derived concentrations used in the experiments from concentrations known to induce mortality in other studies, as available (Post 1979; Smart et al. 1979; Yoshikawa et al. 1988a, 1988b, 1989; Gelwicks et al. 1998). We supplied controls with source water (i.e., BFTC spring water) with ambient levels of dissolved CO₂ that we measured at approximately 30 mg CO₂/L. We exposed RBT and WCT to the following six treatments of total CO₂: control, 75, 150, 180, 225, and 275 mg CO₂/L, and applied the following seven treatments to CF and CC: control, 75, 150, 225, 300, 380, and 495 mg CO₂/L.

To achieve target concentrations, we injected CO₂ into a 2.4-m-tall down-flow bubble contactor made in-house and mixed with BFTC spring water (Figure 1). We then added the CO₂-enriched water to experimental tanks at a rate of 9.5 L/min. Experimental tanks used for the CC and CF trials were made of fiberglass, measured 45.7 cm diameter, and held 56.8 L of water. Water exchange in these tanks was at a rate of 7.6 L/min for a total of eight exchanges per hour. Experimental tanks used for the WCT and RBT trials were made of fiberglass, measured 61 cm diameter, and held 102.2 L of water. Water exchange in these tanks was at a rate of 8.3 L/min for a total of 4.4 exchanges per hour. Each experimental tank was equipped with a headspace unit (Figure 1; please see methods in Watten et al. 2004) that made it possible to monitor CO₂ concentrations in each tank with a portable AC/DC infrared carbon dioxide analyzer (GD-888, CEA Instruments, Inc., Westwood, NJ). We adjusted the rate at which we injected CO₂ into the down-flow bubble contactor using a separate flow meter to achieve desired treatment levels. Upon reaching a particular treatment level (based on meter readings), we closed a valve on the CO₂ injection line (preventing CO₂ from entering the down-flow bubble contactor until the setting on the flow meter was changed) and drained the experimental tanks. Trials began after refilling the drained experimental tanks.

For each species, we conducted two separate exposure trials consecutively (i.e., trial 1 on one week followed by trial 2 on the following week). We exposed fish in the experimental tanks to each concentration of CO₂ for a period of 24 h. Because of tank space limitations, we could not expose fish to all concentrations concurrently. Instead, we randomly assigned a maximum of two concentrations to apply on each day of the trial. At the start of each exposure (i.e., 24-h period), we netted 10 fish from all locations in the tank and randomly allocated them with a random number generator to each experimental tank. We then randomly assigned CO₂ concentrations to each tank. We documented total mortality in each tank at the end of each trial and assessed it by lack of response to gentle prodding and lack of opercular movement. We used three replicate tanks per concentration per trial. This experimental design resulted in each treatment level being replicated six times over the course of the two trials per species (3 tanks/trial × 2 trials/species × 10 fish/tank = 60 fish/concentration). Because we conducted trials in a laboratory facility with limitations on space, we were only able to replicate the 380 mg CO₂/L treatment three times for both CF and CC.

We deviated slightly from standard acute toxicity procedures (American Society for Testing and Materials) by not placing fish directly into dilution water containing CO₂ at the test concentration and allowing a short acclimation period (30 min). We did this because we did not want the physical act of fish allocation to the experimental tanks to alter CO₂ concentrations in the tank and to ensure that we achieved the same exposure period for each group of fish. We therefore did not initiate addition of CO₂ to the experimental tanks until 30 min after we transferred the last fish.

We documented pretreatment and posttreatment (i.e., at the end of the treatment) dissolved oxygen concentrations (mg DO/L), water temperatures (°C), and pH with a YSI Professional Plus Quatro handheld water quality meter (YSI, Inc., Yellow Springs, OH; DO, T, pH). At the end of the 24-h exposure, we documented mortality in each tank, and euthanized any live individuals with an overdose of tricaine methanesulfonate (MS-222). Subsequently, we recorded total length (nearest millimeter) and weight (nearest 0.10 g) for each individual.

We developed a Bayesian hierarchical model to explore the relationship between CO₂ concentration and mortality. The model accounts for several important design features. First, the basic experimental unit is the tank of fish, and the response variable is the number of fish dead within a tank. The observed mortality numbers are observations of a binomial random variable at the tank level, and the sample size is the total number of experimental tanks. We plotted the mortality proportions for each species and CO₂ concentration to visualize the data at the tank level (Figure 2).

• y_i is the number of fish dead in the ith experiment;

• p_i is the probability of mortality in the ith experiment;

• logit (y_i) = is the log odds of mortality for the ith experiment, log (p_i/1 – p_i);

• s_i indexes the species of the ith experiment where s ∈ 1, 2, 3, 4;

• r_i indexes the trial of the ith experiment where r ∈ 1, 2;

• t_i indexes the physical tank of the ith experiment where t ∈ 1, 2, 3, 4, 5, 6;

• a_i [r_i] indexes the day within trial of the ith experiment where a ∈ 1, 2, …, 24;

• dose_i is the CO₂ concentration of the ith experiment, and

• weight_i is the average fish weight of the ith experiment.

We did not include length as a predictor in the model due to collinearity with the weight variable. A strong linear relationship is observed between fish weight and length (r = 0.83), and we assumed the effects on mortality due to changes in fish size to be captured by the weight variable (Figure S1, Supplemental Material). With weight as a predictor, we can interpret the relationship between concentration and dose after accounting for the effects of weight.

We took a Bayesian modeling approach because it has several advantages over a mixed effects model for this design. We needed a hierarchical model due to the nesting of the design variable of day within trial, and the Bayesian framework allows for a straightforward and transparent implementation of a hierarchical model. Determining the LC₅₀ and LC₉₅ values for each species was our primary interest, and Bayesian modeling also yields estimates of uncertainty that we could easily obtain by generating posterior draws for these quantities in each step of the Markov Chain Monte Carlo algorithm implemented to obtain posterior draws for model parameters. We calculated the posterior draws for the LC₅₀ and LC₉₅ values by inverting the logistic regression model and solving for dose concentration when the probabilities are 0.5 and 0.95. By contrast, a non-Bayesian modeling approach uses less intuitive methods to obtain associated uncertainties, such as the delta method or Fieller's method used by ED() function within the drc R package (Ritz et al. 2015; R Core Team 2018). Lastly, interpretation of results from a Bayesian model can be more intuitive than interpretation of frequentist analyses.

The model does make several assumptions that require assessment. Since it treats CO₂ concentration as a continuous measure, it assumes that there is a linear relationship between CO₂ concentration and the log odds of mortality. We found this assumption to be adequately met (see Figure 3). Another key assumption in Bayesian hierarchical models is that of exchangeability. The model assumes exchangeability of the tank mortality proportions after accounting for species, tank, trial, day, and weight, as well exchangeability of the coefficients within a batch in their prior distribution. Exchangeability is a less strict assumption than that of independence, assuming that the joint probability density is unchanged with permutations of the indices (Gelman et al. 2014). The batch of coefficients is modeled exchangeability; it is assumed that there is no prior knowledge regarding the dose response relationship for the different fish species.

We fit the Bayesian model with Stan version 2.18.0 (Stan Development Team 2018a) and implemented it through R with the rstan library version 2.17.3 (Stan Development Team 2018b). We used redundant parameterization to speed convergence, and based starting values on results from an initial mixed effects model fit. The Stan model code as well as the R code is included as Text S1 (Supplemental Material). We ran four chains, each with 20,000 iterations, 10,000 of which we discarded as warmup. We examined traceplots and investigated convergence criteria and effective sample size to evaluate model convergence. All values were less than 0.003 different from 1, and all effective sample sizes were greater than 1,378. We examined traceplots for all model parameters, and the traceplots exhibited good mixing of all chains (Figure 4).

The overall average weights of fish used in the trials were 6.7 ± 0.2 g for WCT, 5 ± 0.1 g for RBT, 3.5 ± 0.1 g for CC 6.7 ± 0.2 g, and 6.9 ± 0.1 g for CF. The overall average total lengths were 91 ± 1 mm for WCT, 75 ± 1 mm for RBT, 70 ± 1 mm for CC, and 97 ± 1 mm for CF. Mean water quality parameters remained within physiologically tolerable ranges (e.g., > 6 mg DO/L for cold water species) in all treatments throughout all trials. However, there was a decline in pH between pretreatment and posttreatment readings except in the control group (Table 1). In comparisons among treatments, posttreatment pH was lower in treatments with higher levels of CO₂, and average pH varied at maximum by 1.22 pH units between control and 495 mg CO₂/L treatments in the CF trials.

Figure 2 displays the estimated dose mortality curves. The error bounds shown display the 2.5th and 97.5th percentiles of the posterior draws, representing uncertainty in the curves. We found RBT and WCT to have steeper dose response curves than CC and CF, with evidence of a positive difference in slopes for the RBT–CC, WCT–CC, RBT–CF, and WCT–CF pairwise comparisons. There was little evidence for a difference in slopes for the CF–CC and WCT–RBT comparisons (Table 2). We also found WCT and RBT to have lower LC₅₀ and LC₉₅ values than CC and CF. Figures 5 and 6, respectively, display posterior draws for the LC₉₅ and LC₅₀ quantities. The 24-h LC₅₀ (95% posterior interval) estimates were 193.5 (177.6–212.0) mg CO₂/L for WCT, 150.7 (138.2–166.0) mg CO₂/L for RBT, 340.4 (308.7–375.4) mg CO₂/L for CC, and 352.2 (313.9–391.1) mg CO₂/L for CF. We estimated the 24-h LC₉₅ (95% posterior interval) values as 245.0 (222.2–272.2) mg CO₂/L for WCT, 190.6 (177.2–207.8) mg CO₂/L for RBT, 422.5 (374.7–474.5) mg CO₂/L for CC, and 434.2 (377.2–492.2) mg CO₂/L for CF. We did not observe mortality in WCT and RBT below 150 mg CO₂/L, and did not observe mortality in CC and CF below 225 mg CO₂/L. We observed complete mortality (i.e., all fish in all tanks) at 275 mg CO₂/L (38,672 microatmospheres [μatm]), 225 mg CO₂/L (30,711 μatm), and 495 mg CO₂/L (65,708 μatm for CC; 77,213 μatm for CF) for WCT, RBT, and both CC and CF, respectively.

Carbon dioxide induced mortality at multiple concentrations, and odds of mortality increased as concentration increased for all four species. The lowest experimental concentration inducing mortality in fingerlings was 150 mg CO₂/L for WCT and RBT and 225 mg CO₂/L for CC and CF. The estimated slopes relating the probability of mortality to dose were similar between WCT and RBT, and between CC and CF, though there was no statistical evidence of a difference among slopes defining the dose-response curves for each species (Table 2). However, we found the 24-h LC₉₅ values to differ among cold water–adapted species and warm water–adapted species groups, with statistical evidence for all pairwise differences in 24-h LC₉₅ values, except between CC and CF. The estimated LC₉₅ values of WCT and RBT are 245.0 (222.2–272.2) and 190.6 (177.2–207.8) mg CO₂/L, respectively, and the values are almost half the estimated 24-h LC₉₅ values for CC and CF, which were 422.5 (374.7–474.5) and 434.2 (377.2–492.2) mg CO₂/L, respectively.

Previous research has documented species-specific sensitivities to environmental stressors. At a very basic level, tolerance to changes in water chemistry can vary among species. Northern Pike Esox lucius, for example, have greater tolerance of anoxia (e.g., dissolved oxygen levels ranging from 0.20 to 0.30 mg/L) than Bluegill Lepomis macrochirus and Largemouth Bass Micropterus salmoides (exhibit stress at levels ≤ 0.60 mg/L; Cooper and Washburn 1949). Fingerling Shortnose Sturgeon Acipenser brevirostrum exposed to un-ionized ammonia experienced mortality at higher concentrations (0.58 mg/L) and were less sensitive than post-swim-up larval Colorado Pikeminnow Ptychocheilus lucius, Razorback Sucker Xyrauchen texanus, and Fathead Minnow Pimephales promelas (Fontenot et al. 1998; Fairchild et al. 2002). Aquatic species also have varying susceptibilities to introduced chemicals, such as piscicides. Some species of fish (e.g., Slimy Sculpin Cottus cognatus) are more susceptible than others (e.g., RBT and Brook Trout Salvelinus fontinalis) to the negative effects of rotenone (Willis and Ling 2000; Grisak et al. 2007).

The similarities in mortality between CC and CF and between RBT and WCT are not surprising given the similar thermal-chemical characteristics present in the preferred habitats of the warm water–adapted and cold water–adapted species pairs (Mayden 1992; Stearley and Smith 1993; Behnke 2002). Our findings suggest that the selectivity of CO₂ as a pesticide may ultimately be limited if one is trying to remove a species that is closely related to a nontarget native fish (e.g., RBT and WCT). Similarly, if the management goal is to remove a species that has evolved to thrive in habitats naturally characterized by high levels of CO₂ and lower pH levels (e.g., CC or CF), our results indicate that the amount of CO₂ needed to achieve complete mortality (i.e., 495 mg CO₂/L) would also likely result in sublethal effects (i.e., loss of equilibrium) or complete mortality of nontarget native species (Hasler et al. 2016). For example, recent research on the effects of CO₂ on native mussels demonstrated that prolonged exposure (i.e., 28 d) to levels of CO₂ suggested for aquatic invasive species control (24,000–96,000 μatm) resulted in reductions in growth and survival (Waller et al. 2019). The long-term implications of short-term exposure to CO₂ remain unknown. In this regard, use of CO₂ may not ultimately spare nontarget organisms in all application settings, but other characteristics (i.e., being naturally occurring and being readily taken up by primary aquatic producers) make CO₂ an interesting pesticide to consider for aquatic invasive species removal projects. Currently, with the USEPA registration of antimycin-A lapsing, CO₂ and rotenone are the only two general pesticides approved by the USEPA for control of nuisance fish populations. As has been done with rotenone, research identifying lethal concentrations for nontarget species would help fisheries managers planning eradication projects.

Control of nuisance fish populations is not limited to removal, but can also include deterrence and behavior modification (Treanor et al. 2017). Researchers have demonstrated CO₂ to be highly effective in deterring the physical movements of nonnative fishes (Hasler et al. 2019; Suski et al. 2019). Specifically, initial research has shown that CO₂ is effective at deterring passage and reducing swimming velocity in invasive carp species and CF (Donaldson et al. 2016; Cupp et al. 2017a). One could even postulate the future integration of deterrence with lethal control where CO₂ is used to direct fish to a removal or sorting site where fisheries managers could retain native species and dispatch target species.

Despite the successful determination of lethal CO₂ concentrations for all species included in this study, several factors qualify our findings. First, we evaluated CO₂ effects using a single water temperature in our experiments and recent work indicated that temperature may influence avoidance behaviors to CO₂ treatments and that temperature-related effects may be species-specific (Tix et al. 2018). Other work suggested that increased water temperatures can exacerbate the lethal effects of increased CO₂ concentrations (Gelwicks et al. 1998). Future users should recognize the potential synergistic effects of water temperature (and other water quality parameters, like alkalinity) and CO₂ when deciding on application concentrations (Treanor et al. 2017). Additionally, the concentrations used did not represent a geometric series. For example, for each species tested, there were gaps between a concentration demonstrated to induce zero mortality and one that induced 60% mortality, as required by protocol. This makes it difficult to definitively determine the concentrations that would induce a particular percentage of mortality (i.e., LC₅₀). There were also instances in which mortality at a certain concentration varied across tanks (Figure 2). This potentially demonstrates individual tolerance of conspecifics to CO₂ at that treatment, and suggests there may be phenotypic plasticity to CO₂ as has been shown for temperature (Yu et al. 2018) and other environmental stressors. Although there was evidence of differences among pretreatment and posttreatment water quality parameters, we do not believe that these differences directly contributed to the observed mortality rates. We expect the differences between pretreatment and posttreatment pH levels in all groups exposed to additions of CO₂ given the known effects of CO₂ on pH (i.e., increases in CO₂ result in a decrease in pH; Colt et al. 2012). And, because pH never dropped to levels known to affect survival of RBT (Wagner et al. 1997), CC (Edwards and Twomey 1982), and CF (Mischke and Wise 2008; Mischke and Chatakondi 2012), we believe that mortality occurred principally as a result of CO₂ treatment and not the anticipated increases in acidification (Treanor et al. 2017). We used juveniles in our study, but field applications would likely affect individuals of a range of sizes. Initial research has shown that CO₂ negatively impacted both juvenile and adult fishes at levels within the range used in our trials (Treanor et al. 2017).

Although CO₂ may not be suitable for selective removal of some invasive fish species (e.g., RBT), CO₂ applications in appropriate settings have the potential to serve in conjunction with or as an alternative to traditional piscicides (e.g., rotenone). This work represents a necessary addition to a growing body of literature formally evaluating the potential for CO₂ to be used in fisheries management for the exclusion or removal of nonnative fish in both flowing (e.g., Hasler et al. 2019) and standing (e.g., Cupp et al. 2017a) waters. Additional research is needed to further explore the susceptibility of other species to CO₂, and how susceptibility varies across individuals of the same species (Cupp et al. 2017b; Tucker et al. 2019) or based on the nutritional status of an individual or population (Suski et al. 2019). In addition, implementation requires more work to carefully quantify the range of lethal and nonlethal impacts to nontarget organisms, as well as the abiotic factors (e.g., water temperature, flow velocity, and area and volume of the treated waters) that may influence the efficacy of a CO₂ application in a natural setting. With increased interest in CO₂ in invasive species management, CO₂ delivery systems are diversifying through experimentation and include gaseous and solid-state (Cupp et al. 2018) approaches. Not all applications will be appropriate for each field setting, and further development of application tools will only help to expand the potential for use of CO₂ in fisheries management. Ultimately, it is possible that fisheries managers could use CO₂ either alone or in concert with other mechanical removal techniques to eliminate nonnative aquatic species.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author for the article.

Text S1. Stan code for Bayesian model and rstan code for model setup and run. From 2012 to 2013, we exposed Common Carp Cyprinus Carpio, Channel Catfish Ictalurus punctatus, Rainbow Trout Oncorhynchus mykiss, and Westslope Cutthroat Trout Oncorhynchus clarkii lewisi under laboratory conditions to each of six or seven CO₂ concentrations at the Bozeman Fish Technology Center for a period of 24 h.

Found at DOI: https://doi.org/10.3996/JFWM-20-012.S1 (60 KB DOCX).

Figure S1. Residual plots used to check assumptions of the linear fish weight model. From 2012 to 2013, we exposed Common Carp Cyprinus Carpio, Channel Catfish Ictalurus punctatus, Rainbow Trout Oncorhynchus mykiss, and Westslope Cutthroat Trout Oncorhynchus clarkii lewisi under laboratory conditions to each of six or seven CO₂ concentrations at the Bozeman Fish Technology Center for a period of 24 h.

Found at DOI: https://doi.org/10.3996/JFWM-20-012.S2 (11.16 MB DOCX).

Reference S1.Edwards EA, Twomey KA. 1982. Habitat suitability index models: common carp. U. S. Department of Interior, U.S. Fish and Wildlife Service. FWS/OBS-82/10.12

Found at DOI: https://doi.org/10.3996/JFWM-20-012.S3 (2.28 MB PDF).

Reference S2.Fairchild JF, Allert AL, Poulton BC, Graham RV. 2002. A site-specific assessment for the impacts of ammonia on Colorado pikeminnow and razorback sucker populations in the Upper Colorado River, adjacent to the Atlas Mill Tailings Pile, Utah. U.S. Geological Survey, Columbia Environmental Research Center, Final Report to the U.S. Fish and Wildlife Service, Off-Refuge Contaminant Assessment Program (91435).

Found at DOI: https://doi.org/10.3996/JFWM-20-012.S4 (2.18 KB PDF).

This research was funded by the United States Department of Agriculture–North Central Regional Aquaculture Center. Joanne Grady (USFWS–Region 6 Aquatic Invasive Species Coordinator) provided funds for the CO₂ monitoring equipment. We thank Ken Staigmiller (Montana Fish, Wildlife & Parks) for his efforts in the collection of the CC used in these experiments. We also would like to thank the Aquatic Animal Drug Approval Partnership Program for providing the CF used in this research. Additional thanks to the USFWS and Montana State University staff that helped to complete this work, specifically Lynn DiGennaro, Matt Toner, Luke Holmquist, Mariah Talbott, Michael Forzley, Paige Maskill, and Cal Fraser. The Associate Editor and journal reviewers provided invaluable feedback and greatly improved the manuscript. Finally, we would like to thank Richard Erickson (U.S. Geological Survey) for initial statistical review and Megan Higgs (Neptune and Company) for assistance with statistical analyses. The methodology used in this research was approved by the Montana State University Institutional Animal Care and Use Committee.

Any use of trade, product, website, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.