The interactive effects of predator stress, predation, and the herbicide Roundup

Transcription

1 The interactive effects of predator stress, predation, and the herbicide Roundup RICK A. RELYEA Department of Biological Sciences, Darrin Fresh Water Institute, Rensselaer Polytechnic Institute, Troy, New York USA Citation: Relyea, R. A The interactive effects of predator stress, predation, and the herbicide Roundup. Ecosphere 9(11):e /ecs Abstract. As the number of studies examining the effects of contaminants grows, ecologists are becoming increasingly aware that contaminants can interact with natural stressors (e.g., competition and predator cues) in their effects on nontarget animals. In amphibians, predator cues can make contaminants more lethal under laboratory conditions, but the opposite outcome has been observed under more natural conditions with stratified water columns; stratification causes more pesticide to be present near the surface while predator cues scare spring-breeding amphibians down to the benthos. I examined whether this phenomenon also occurs in three species of summer-breeding amphibians (Hyla versicolor, Rana clamitans, and Rana catesbeiana) that were raised in outdoor mesocosms. Specifically, I asked how amphibian survival was affected by multiple concentrations of a common herbicide (glyphosate; commercial name: Roundup), the herbicide combined with chemical cues from predators (caged larval dragonflies; Anax junius), and the herbicide combined with lethal predators. Environmentally relevant concentrations of the herbicide caused high rates of tadpole mortality, but this outcome was substantially reversed by the addition of predator cues. With lethal predators, the tadpoles experienced such high mortality that the herbicide caused no additional effect. Roundup also induced morphological changes in Hyla versicolor, and the induced traits were different from those induced by predators. Collectively, these results suggest that while predator cues can make pesticides less lethal when thermal stratification occurs, highly lethal predators can overwhelm these effects. Thus, the impacts of such contaminants can be dramatically different in environments that do or do not contain high-risk predators. Key words: bullfrogs; ecotoxicology; gray tree frogs; green frogs; phenotypic plasticity. Received 10 September 2018; accepted 17 September Corresponding Editor: Debra P. C. Peters. Copyright: 2018 The Author. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. relyer@rpi.edu INTRODUCTION When considering the effects of contaminants on biodiversity, the current paradigm is to conduct highly controlled, short-term experiments on a few species of model organisms in the laboratory and use the results of these experiments to estimate allowable concentrations in nature that will not cause harm to plants and animals. During the past decade, however, there has been a growing appreciation that natural stressors can fundamentally alter the outcomes of exposures to contaminants (Relyea and Hoverman 2006, Clements and Rohr 2009). This poses a major challenge to regulatory agencies, since the impacts of contaminants can be context-dependent. In response, ecologists and toxicologists are striving to provide guidance by developing general rules of thumb for the ecological conditions under which contaminants can become more or less harmful to nontarget species. One clear example of context dependence for contaminant impacts is the effect of pesticides in the presence and absence of predator cues. Predator cues can be visual, tactile, or chemical (from both the predator and the consumed prey), 1 November 2018 Volume 9(11) Article e02476

2 and the cues induce changes in prey physiology, behavior, and morphology in ways that typically reduce the prey s probability of being detected, captured, or consumed (Lima 2002, Miner et al. 2005, Ohgushi et al. 2012). However, predator cues also can interact with pesticide effects in a variety of ways. In the first experiments under laboratory conditions, researchers discovered that the chemical cues from aquatic predators could make the pesticide carbaryl more lethal to gray tree frog tadpoles (Hyla versicolor; Relyea and Mills 2001). Subsequent experiments confirmed that this phenomenon occurred in other species of tadpoles and with other pesticides, including the insecticide malathion and the herbicide glyphosate. In fact, the presence of predator cues can cause some pesticides to become nearly 50 times more lethal, although not all species exhibit synergistic interactions (Relyea 2003, 2004b, 2005c, Giri et al. 2012, Yadav et al. 2013). Subsequent studies on zooplankton and aquatic insects have found that the nature of the predator pesticide interaction varies among species and among pesticides that possess different modes of action (Qin et al. 2011, Trekels et al. 2011). While the stress of predator cues can make pesticides more lethal under laboratory conditions, it is important to assess whether similar outcomes occur under more natural conditions. In an example with tadpoles raised in outdoor mesocosms, Relyea (2012) discovered that predator cues from caged dragonflies and caged salamanders caused the herbicide glyphosate (commercial name: Roundup; Monsanto Corporation, St. Louis, Missouri, USA) to become less lethal to an assemblage of spring-breeding amphibians: larval wood frogs (Rana sylvatica), leopard frogs (Rana pipiens), and American toads (Anaxyrus americanus). This outcome was surprising given that it is opposite from the results observed under laboratory conditions. The underlying cause of this reduced mortality was that in outdoor mesocosms, the water column naturally experiences temperature stratification, which inhibits the mixing of the water and thereby causes higher herbicide concentrations to occur near the surface than near the bottom. The addition of predators in floating cages caused the tadpoles to adaptively respond by avoiding the surface waters and spending most of their time near the bottom. Because less herbicide existed near the bottom due to stratification, tadpole mortality was reduced. This study confirmed that additional ecological conditions such as water stratification can play a key role in determining toxicity. However, we clearly need many more studies on this phenomenon before we can arrive at any generalities. In assessing the interactive effects of predators and pesticides, it is important to not only quantify the impact of predator cues on prey, but also directly compare these effects to lethal (i.e., uncaged) predators that emit cues while simultaneously consuming the prey. In doing so, we can determine whether the mortality caused by the pesticide and predator cues is overwhelmed by actual predation. I am not aware of any studies that have provided such a comparison. Several studies have examined the impact of pesticides of predation rates under concentrations that are sublethal to prey and alter either predator or prey behavior (Dodson et al. 1995, Broomhall 2002, 2004, Relyea and Edwards 2010), but few have examined predation effects under pesticide concentrations that are lethal to prey (Boone and Semlitsch 2001, Hanlon and Relyea 2013). In addition to affecting survival, predators and pesticides can also induce changes in animal phenotypes. It is well established that predators can induce anti-predator phenotypes (Lima 2002), but it is less appreciated that pesticides can do so as well. In zooplankton, for example, pesticides can induce or inhibit morphological defenses that are associated with predator defense (Hanazato 1991, 1999, Barry 1998, 1999, 2000). Pesticides can also induce morphological changes in vertebrates. For example, Relyea (2012) discovered that the herbicide glyphosate can induce changes in the morphology of two tadpole species (wood frogs and leopard frogs) and that these changes are strikingly similar to the morphological changes induced by predators in a wide range of amphibian species (Van Buskirk 2002). Moreover, when the herbicide and predator cues were both present, the effects were additive, such that even larger morphological defenses were produced. While these results were novel, it appears that there have been no subsequent studies of herbicide-induced morphology, which precludes us from knowing how common such responses may be. 2 November 2018 Volume 9(11) Article e02476

3 I addressed these challenges in an outdoor mesocosm experiment that exposed an assemblage of the tadpoles from summer-breeding amphibian species to the separate and combined effects of a range of glyphosate concentrations in the presence of no predators, predator cues, and lethal predators. Tadpoles are an excellent model system because a large amount of data exists on their responses to glyphosate (Relyea 2005c, Bernal et al. 2009, Jones et al. 2009) and their responses to predators (Laurila 2000, Van Buskirk 2002, Relyea 2003, 2004a). METHODS Pesticide background Glyphosate is the number one herbicide in the world and is sold under numerous commercial names by multiple manufacturers, including multiple formulations of Roundup. It is used by farmers, residents, and industries to control undesired plant growth and is an important component in growing Roundup-Ready crops. Glyphosate inhibits the synthesis of key amino acids in plants, but typically requires the addition of a surfactant to allow the active ingredient to penetrate the waxy outer layer of plant leaves. Commercial formulations for terrestrial applications typically contain the surfactant POEA (polyethoxylated tallow amine), although other formulations contain surfactants that are held as trade secrets. Regardless, multiple formulations containing surfactants (which appear to be the primary cause of amphibian and fish mortality) cause similar levels of high toxicity to fish and larval amphibians at environmentally relevant concentrations (Mann and Bidwell 1999, Relyea 2005a,b,c, 2006, 2011). In regard to crops, the application timing of glyphosate commonly coincides with the movement of amphibians across agricultural fields (Berger et al. 2013). When examining the impacts of pesticides on ecological communities, it is important that we examine the range of ecologically relevant concentrations. As Wagner et al. s (2013) review makes clear, there is currently a paucity of field data (see also Smalling et al. 2015). The amount of glyphosate entering aquatic habitats depends on application rates, interception by vegetation, and whether the applications are direct or indirect. Past studies have provided estimated worst-case scenarios of mg a.e./l (i.e., acid equivalents) while observed worst-case scenarios range from 1.7 to 5.2 mg a.e./l (Edwards et al. 1980, Boutin et al. 1995, Mann and Bidwell 1999, Giesy et al. 2000, Solomon and Thompson 2003, Thompson et al. 2004). The half-life of glyphosate varies with environmental conditions, but ranges from 8 to 120 d (Barolo 1993). The mesocosm experiment To examine the effects of Roundup and predators on an amphibian community, I conducted a mesocosm experiment that employed a completely randomized design with a factorial combination of four nominal Roundup concentrations (0, 1, 2, or 3 mg a.e./l of glyphosate) crossed with three predator treatments (no predator, caged predators, or lethal predators). These 12 treatment combinations were replicated four times for a total of 48 experimental units. The experimental units were 757-L cattle tanks that were filled with approximately 473 L of well water on June On 17 and 20 June, respectively, I added 15 g of rabbit chow and 200 g of leaf litter (Quercus spp.) to each tank to serve as initial nutrient sources for the community and to serve as refuge from the predators. On 20 June, I collected pond water from four ponds that contained zooplankton, phytoplankton, and periphyton and added an aliquot to each tank after screening the water for predators. Soil was not included in the mesocosms, but past work has confirmed that the presence of soil does not affect the toxicity of Roundup to amphibians living in mesocosms (Relyea 2005b). All mesocosms were covered with 65% shade cloth lids to prevent other organisms from ovipositing in the experiment. After waiting 18 d for the mesocosms to develop algal and zooplankton communities, I added three species of amphibians to each tank. The amphibians were originally collected as newly oviposited egg masses in natural ponds (5 bullfrog egg masses, 16 green frog egg masses, and 10 gray tree frog egg masses). Egg masses were hatched in wading pools filled with aged well water until they were large enough to be safely handled. On 8 July (defined as day 0 of the experiment), I added 30 tadpoles of each species for a total of 90 tadpoles per mesocosm (initial mean mass 1SE: 3 November 2018 Volume 9(11) Article e02476

4 bullfrogs = 25 2 mg, green frogs = 30 2, and gray tree frogs = 43 4mg). This density of hatchling tadpoles (22 tadpoles/m 2 for each species) is within the range of densities observed in nature for these species (R.A. Relyea, unpublished data). On day 1, I added two floating predator cages to each tank and then applied the three predator treatments. The cages were constructed from 236-mL plastic cups with a screen over the opening held on by a rubber band. For the no-predator treatments, the cages remained empty. For the caged-predator treatment, each cage held a larval dragonfly (Anax junius) that was fed approximately 300 mg of larval gray tree frogs and green frogs three times per week. Previous work has found that caged predators emit waterborne cues and that these chemical cues induce changes in tadpole behavior and morphology (Kats and Dill 1998, Laurila 2000, Relyea 2001). The dragonfly diet did not include bullfrog tadpoles since prior research (Schoeppner and Relyea 2005) has shown that tadpoles respond similarly to predators eating congeneric species of tadpoles (i.e., larval green frogs and bullfrogs). When the caged predators were fed, I also lifted the empty cages of the other treatments to equalize disturbance across all mesocosms. The lethalpredator treatment consisted of two larval dragonflies that were added to the mesocosms as caged predators for 2 d and then released to permit the tadpoles an opportunity to be aware that predators were in their environment and respond accordingly. Once the predators were released, I left the empty cages in the experiment. On the same day that the lethal predators were released (day 3), I applied the glyphosate treatments. I used a popular commercial formulation (Roundup Original Max; EPA Registration No ; Monsanto, St. Louis, Missori, USA) that contains both glyphosate (reported to contain 540 g a.e./l) and a surfactant that helps glyphosate penetrate the waxy cuticle of plant leaves. The composition of Roundup surfactants is treated as trade secrets (S. Mortensen, Monsanto, personal communication). However, past work has shown that the toxicity of Roundup Original Max to larval amphibians is nearly identical to glyphosate formulations that contain the popular surfactant POEA (Relyea 2005a,b,c, Relyea and Jones 2009). To achieve the nominal concentrations of 0, 1, 2, and 3 mg a.e./l of glyphosate from this formulation, I added 0, 0.716, 1.636, or ml of Roundup Original Max to the appropriate mesocosms. Each amount was initially dissolved into 300 ml of well water, and then, this water was spread evenly over the surface of the mesocosm to simulate surface runoff, aerial drift, or aerial overspray (which happens in small wetlands during forestry applications; Thompson et al. 2004). Ninety minutes after dosing, I collected water near the surface from all tanks and pooled all samples from a particular glyphosate treatment. This sample was frozen and later shipped to the Mississippi State Chemical Lab (Mississippi, Mississippi State, USA) where the concentration was tested using high-pressure liquid chromatography. The actual concentrations were 1.7, 2.7, and 4.4 mg a.e./l. While these values are 35 70% higher than the nominal values, our subsequent experiments clarified that Roundup stratifies in these mesocosms. For example, Jones et al. (2010) also used nominal concentrations of 1, 2, and 3 mg a.e./l. In that study, the surface water had actual concentrations of 1.5, 3.1, and 4.8 mg a.e./l while the deep water had concentrations of 0.4, 0.81, and 1.3 mg a.e./l. Averaged between deep and shallow water, Jones et al. (2010) had actual concentrations of 0.92, 1.96, and 3.05, which is very close to the nominal concentrations. Moreover, subsequent testing on day 7 and day 14 of that study confirmed that the herbicide concentrations remained stratified for at least 2 weeks. A very similar observation of stratified glyphosate was observed in Jones et al. (2011). Given that these other mesocosm experiments were conducted in the same location and under the same conditions as the current study, it suggests that the tested water samples in the current experiment were stratified and if I had taken samples at the top and bottom of the water column, I would have observed mean concentrations that were very similar to the nominal concentrations. Such stratification of pesticides also has been observed in natural wetlands (Sudo et al. 2004, Ma et al. 2008). On days 5 and 10, I measured the ph and temperature of all tanks. Measurements were taken with a digital meter that was calibrated earlier that day. On day 5, mean temperature of the 12 treatments ranged from 30.2 to 30.8 C and mean 4 November 2018 Volume 9(11) Article e02476

5 ph ranged from 8.3 to 8.5. On day 10, mean temperature of the 12 treatments ranged from 29.0 to 29.4 C and mean ph again ranged from 8.3 to 8.5. Preliminary analyses confirmed that the treatments had no effect on water temperature or ph (results not shown). Because the gray tree frogs grew rapidly and were approaching metamorphosis by day 17, I terminated the experiment on day 17. On this day, all water and leaves were removed from the mesocosms and held in other tanks until complete breakdown the following spring. The surviving tadpoles were euthanized in 2% MS-222 and then preserved in 10% formalin. At a later date, the preserved animals were removed, counted, and weighed to determine survival and individual mass of each species. The proportion of each species surviving in a mesocosm and the mean individual mass of each species in a mesocosm were our amphibian response variables. The larval dragonflies also experienced high survival; I recovered 94% lethal dragonflies and 97% caged dragonflies. Analysis of survival in all three tadpole species I conducted analyses of the survival and growth of all three tadpole species as well as an analysis of relative morphology and tail color intensity for the morphologically plastic gray tree frog tadpoles. For tadpole survival, I conducted a MANOVA on the survival of the three tadpole species. To improve the data fit to the parametric assumptions, I transformed the survival data using an arcsine(square-root) transformation. I conducted subsequent ANOVAs of each species followed by Tukey mean comparison tests. To determine whether the mortality under the various combinations of predators and the herbicide differed from that expected under an assumption of non-interacting effects, I estimated the expected survival based on addition probability (Soluk and Collins 1988, Wissinger and McGrady 1993). In this method, the expected survival from two non-interacting factors is calculated as [P P 9 P H ], where P P and P H are the probabilities of survival from a given predator or pesticide treatment, respectively. I calculate four estimates using the four replicates of each treatment. I then conducted an ANOVA that compared the observed and expected survival values for the six treatments that contained caged or lethal predators crossed with 1, 2, or 3 mg a.e./l of glyphosate. I then conducted Tukey mean comparisons tests. Analysis of mass in all three tadpole species Because some treatments had zero surviving gray tree frog tadpoles, I analyzed individual tadpole mass using separate ANOVAs for each species. To better meet the parametric assumptions, I log-transformed the mass of green frog tadpoles. For those response variables with pesticide-by-predator interactions, I examined the impacts of the increasing pesticide concentrations within each predator treatment and compared the three pesticide additions against the control using a Dunnett s test. Finally, to assess the lethal concentration required to cause 50% mortality (i.e., the LC50), I conducted probit analyses. Analysis of morphology in gray tree frogs Of the three tadpoles species used in the experiment, only the gray tree frog is known to be morphologically plastic in response to predator cues (Relyea 2001). To determine whether gray tree frogs altered their morphology in response to the predator cues and herbicide concentrations, I measured several morphological dimensions, made the dimensions mass-adjusted, and then conducted a MANOVA on the mass-adjusted dimensions. Since few gray tree frog tadpoles survived in the lethal-predator treatment, I only examined morphological changes using the nopredator and caged-predator treatments (crossed with all four herbicide treatments). I measured all tadpoles available from each mesocosm, with a mean of 21 tadpoles per mesocosm. I initiated the morphological analysis focusing on the five tadpole traits that are commonly measured in studies of tadpole plasticity: tail depth and length; body depth, length, and width (see Fig. 1 in Relyea 2000). I log-transformed the linear dimensions and mass of each tadpole to improve the linearity of the relationships between mass and each linear dimension and then calculated the mean masses and dimensions for each mesocosm. Using these mesocosm means, I conducted a MANCOVA with predator treatments, herbicide treatments, their interaction, and mass as a covariate. Once I confirmed that the treatments contained parallel relationships (i.e., no significant 5 November 2018 Volume 9(11) Article e02476

6 mass-by-treatment interactions), I removed the mass-by-treatment interactions and re-ran the MANCOVA. This approach to produce massindependent morphological traits has been used in many past studies of morphological plasticity (Hoverman and Relyea 2008, 2009). Analysis of color intensity in gray tree frogs To determine the intensity of the red tail that is common to predator-induced gray tree frog tadpoles, I took photographs of the tadpoles from each experimental unit. Given that few tadpoles survived with lethal predators present, I restricted the color analysis to tadpoles exposed to the no-predator and caged-predator treatments (each crossed with the four herbicide treatments). From the tadpole photographs, the tail was cropped and all black spots were removed by deleting the black pixels. From the remaining pixels, I examined the intensity of red (on a scale of 0 256) and then divided that value by the number of pixels in the image to determine the mean pixel intensity for each tadpole. I analyzed color intensity using a mean of 4.9 tadpoles per Fig. 1. Survival of three species of tadpoles when exposed to a factorial combination of predator treatments (no predator, caged predator, lethal predator) crossed with a range of glyphosate concentrations (0 3 mg a.e./l). Data are means 1SE. The predator and herbicide treatments exhibited significant interactions for all three species. Asterisks and crosses indicate glyphosate concentrations that were significantly or marginally different from the control within a given predator treatment (P < 0.05 or P < 0.07, respectively). Green symbols indicate the expected survival in the absence of an interaction between the predator and herbicide treatments; hash tags indicate significant differences between expected and observed survival. 6 November 2018 Volume 9(11) Article e02476

7 experimental unit. I then calculated the mean intensity for each experimental unit. I conducted an ANOVA on all tadpoles from the no-predator and caged-predator treatments (crossed with the four herbicide treatments) since few tadpoles survived the lethal-predator treatment. RESULTS Amphibian survival The MANOVA found significant effects of the treatments and their interaction (Table 1). As a result, I examined the univariate effects for each species. For gray tree frog tadpoles, there were significant effects of predators, pesticides, and their interaction (Table 2). In the absence of predators, increased herbicide concentrations caused a sharp reduction in survival at the highest herbicide concentration, from 90% to 23% (P = 0.001). When caged predators were present, however, increased herbicide concentrations caused only a modest reduction in survival at the highest concentration (from 88% to 65%), which was not quite significant (P = 0.066). In comparing the observed versus expected survival (under an assumption of non-interacting effects), I confirmed that the observed survival was higher than expected when 2 or 3 mg a.e./l of glyphosate was combined with caged-predator cues (all P < 0.05). Table 1. Multivariate test results examining the effects of three predator cue treatments crossed with four concentrations of Roundup Original Max on the survival of gray tree frogs, green frogs, and American bullfrog tadpoles. Multivariate test (Wilks lambda) df F-value P-value Predator 6, <0.001 Herbicide 9, <0.001 Predator 9 Herbicide 18, When lethal predators were present, gray tree frog survival was low regardless of herbicide concentration, averaging about 8% (all P > 0.28). There were no differences between observed and expected survival under when lethal predators were combined with 1, 2, or 3 mg a.e./l of glyphosate (P > 0.05). For gray tree frogs, the probit analyses indicated an LC50 values of 2.3 mg a.e./ L (95% CI = mg a.e./l) without predators. With caged predators, survival remained above 50% across all herbicide concentrations, precluding any estimate of an LC50. For green frog tadpoles, there was a significant effect of predators, a nearly significant effect of pesticides, and a significant interaction (Table 2). In the absence of predators, increasing the herbicide concentrations caused survival to decline from 91% to 62% (P = 0.02). With caged predators, there was no decline in survival as the concentration of the herbicide increased (all P > 0.6). In comparing the observed versus expected survival, I found that the observed survival was higher than expected when 3 mg a.e./l of glyphosate was combined with caged-predator cues (P < 0.05). With lethal predators, only about 60% of the tadpoles survived and increasing the herbicide concentration had no effect compared to the control (P > 0.16). There were no differences between observed and expected survival under when lethal predators were combined with 1, 2, or 3 mg a.e./l of glyphosate (P > 0.05). Given that mortality never exceeded 50%, with or without caged predators, this precluded the ability to estimate LC50 values, although the LC50 value is likely not much greater than 3 mg a.e./l. For bullfrog tadpoles, there were significant effects of predators, pesticides, and their interaction (Table 2). Without predators, bullfrog survival declined from 87% to 47% as the herbicide concentration increased (P = 0.007). With caged Table 2. Univariate test results examining the effects of three predator cue treatments crossed with four concentrations of Roundup Original Max on the survival of gray tree frogs, green frogs, and American bullfrog tadpoles. Univariate tests Predator (df = 2,36) Herbicide (df = 3,36) Predator 9 Herbicide (df = 6,36) Gray tree frog survival <0.001 < Green frog survival < Bullfrog survival < Note: Only P-values are listed. 7 November 2018 Volume 9(11) Article e02476

8 predators present, there was no longer an herbicide effect, with survival ranging from 88% to 93% (all P > 0.37). In comparing the observed versus expected survival, the observed survival was higher than expected when 2 or 3 mg a.e./l of glyphosate was combined with caged-predator cues (P < 0.05). With lethal predators present, bullfrog survival was reduced to about 30%; increasing the concentration of herbicide had no effect compared to the control (all P > 0.58). There were no differences between observed and expected survival under when lethal predators were combined with 1, 2, or 3 mg a.e./l of glyphosate (P > 0.05). The probit analyses indicated an LC50 values of 3.0 mg a.e./l without predators (95% CI = mg a.e./l). Mortality was less than 50% with caged predators, which precluded me from estimating an LC50 with caged predators. Amphibian mass For gray tree frogs (Fig. 2), there was a main effect of predators (F 2,33 = 17.7; P < 0.001), no main effect of herbicide (F 3,33 = 1.3; P = 0.300), and a Fig. 2. Mass of three species of tadpoles when exposed to a factorial combination of predator treatments (no predator, caged predator) crossed with a range of glyphosate concentrations (0 3 mg a.e./l). Data are means 1SE. The gray tree frogs and bullfrogs both experienced significant interactions of the predator and herbicide treatments, whereas green frogs were unaffected by the treatments. Asterisks and crosses indicate glyphosate concentrations that were significantly or marginally different from the control within a given predator treatment (P < 0.05 or P < 0.07, respectively). 8 November 2018 Volume 9(11) Article e02476

9 significant interaction (F 6,33 = 5.4; P < 0.001). Because of the interaction, I examined the herbicide effect within each predator treatment. Without predators, increased herbicide concentrations caused an increase in mass (F 3,12 = 9.9, P = 0.001) that became significantly greater than the control at 3 mg a.e./l (P = 0.001). There was no herbicide effect within the caged-predator (F 3,12 = 1.2, P = 0.351) or lethal-predator (F 3,12 = 2.3, P = 0.148) treatments. I then examined green frog mass (Fig. 2). The ANOVA found no main effect of predators (F 2,36 = 11.9; P = 0.166), no main effect of herbicide (F 3,36 = 1.7; P = 0.191), and no significant interaction (F 6,36 = 1.4; P = 0.237). The ANOVA on bullfrog mass detected no main effect of predators (F 2,36 = 0.2; P = 0.841), a main effect of herbicides (F 3,36 = 7.7; P < 0.001), and an interaction (F 6,36 = 4.9; P = 0.001). Because of the interaction, I examined the herbicide effect within each predator treatment (Fig. 2). Without predators, increasing the herbicide concentration caused an increase in mass (F 3,12 = 4.3, P = 0.027), becoming nearly significant at the highest concentration (P = 0.060). With caged predators, there was no effect of the herbicide (F 3,12 = 1.2, P = 0.340). With lethal Table 3. Multivariate test results examining the effects of two predator cue treatments crossed with four concentrations of Roundup Original Max on the mass-adjusted morphological dimensions of gray tree frog tadpoles. Multivariate test (Wilks lambda) df F-value P-value Mass 5, <0.001 Predator 5, <0.001 Herbicide 15, Predator 9 Herbicide 15, predators, there was an effect of herbicide concentration (F 3,12 = 9.3, P = 0.002); all three herbicide additions caused a decline in bullfrog mass compared to the control (all P < 0.01). Gray tree frog morphology The MANCOVA on the mass-adjusted morphology of gray tree frog tadpoles found a significant effect of predator, herbicide, and the mass covariate, but there was no predator-by-herbicide interaction (Table 3). The subsequent ANCOVAs on individual dimensions found widespread effects of predators but fewer effects of the herbicide (Table 4; Fig. 3). Tadpole tails became relatively deeper with predator cues and there was a marginal effect of herbicide concentrations, but the subsequent Dunnett s test detected no differences between the control and each of the three herbicide concentrations (all P > 0.13). Tadpole tail length was unaffected by predator cues or the herbicide concentrations, although there was a marginal interaction. When I examined the herbicide effects within each predator treatment, I found no effect of the herbicide concentrations on tail length (all P > 0.2). I also examined the three relative body dimensions (Table 4; Fig. 3). Bodies became relatively deeper with predator cues but were unaffected by the herbicide. Body length was unaffected by either factor, but there was a significant interaction. In the absence of caged predators, there was no herbicide effect (F 3,11 = 1.4, P = 0.227); in the presence of caged predators, there was a marginal herbicide effect (F 3,11 = 3.2, P = 0.067). Body width was not affected by predator cues, but it was affected by the herbicide. Compared to the control, adding 1 or 2 mg a.e./l of Roundup caused marginal increases in body width (P = and P = 0.064, respectively). Table 4. Univariate test results examining the effects of two predator cue treatments crossed with four concentrations of Roundup Original Max on the mass-adjusted morphological dimensions of gray tree frog tadpoles. Univariate tests Mass (df = 1,23) Predator (df = 1,23) Herbicide (df = 3,23) Predator 9 Herbicide (df = 3,23) Tail depth <0.001 < Tail length < Body depth <0.001 < Body length < Body width < Note: Only P-values are listed. 9 November 2018 Volume 9(11) Article e02476

10 (F 1,24 = 22.4; P < 0.001), but there was no main effect of the herbicide (F 3,24 = 0.5; P < 0.660) nor an interaction (F 3,24 = 0.9; P = 0.435). DISCUSSION The mesocosm experiment demonstrated that environmentally relevant concentrations of the herbicide Roundup caused typical amounts of mortality across three species of larval amphibians, but this mortality was dramatically reduced in the presence of predator cues. When lethal predators were present, however, mortality was high at all herbicide concentrations. This means that the mortality caused by the herbicide is of similar magnitude to the mortality caused by lethal predators. It also means that the positive effects of predator cues on amphibian survival are completely eliminated by the negative effects of predation on the tadpoles. The effects of the predator and herbicide treatments on mass varied among the three species. Finally, for the morphologically inducible gray tree frog tadpoles, I detected significant changes induced by predator cues but only marginal changes induced by increased concentrations of herbicide; notably, these Roundup-induced changes in morphology did not resemble predator-induced changes in morphology. Fig. 3. Relative morphology (log-transformed) and intensity of the red tail color of gray tree frog tadpoles. All dimensions were mass-adjusted prior to analysis. The predator and herbicide treatments exhibited no significant interactions. Predator cues caused significant increases in relative tail depth and body depth. Data are means 1SE. Crosses indicate glyphosate concentrations that were marginally different (P 0.077) from the control. Caged predators induced tails that were more intensely red, but the herbicide concentrations had no effect. Gray tree frog tail color Finally, I conducted an ANOVA on tail color intensity of gray tree frog tails (Fig. 3). I found predators caused the tails to become more red Effects of the herbicide on amphibian survival Increased concentrations of the herbicide in the absence of predator cues were associated with increased mortality, with significant mortality happening at 3 mg a.e./l for all three amphibian species. There was a concomitant increase in individual tadpole mass, suggesting a release from competition. For two of the species raised without predator cues, there was sufficient mortality to estimate LC50 values, which ranged from 2.3 mg a.e./l in gray tree frog tadpoles to 3.0 mg a.e./l in bullfrog tadpoles. These LC50 values are consistent with a large number of laboratory LC50 studies that have exposed tadpoles to a range of herbicide concentrations for 1 4 d using commercial formulations of glyphosate (e.g., Roundup Original, Roundup Original MAX, Roundup Weathermax, Vision, Cosmo-Flux). Such studies have estimated LC50 values at mg a.e./l (Mann and Bidwell 1999, Lajmanovich et al. 2003, Edginton et al. 2004, Howe et al. 2004, Relyea 10 November 2018 Volume 9(11) Article e02476

11 2005c, Bernal et al. 2009, Relyea and Jones 2009, Dinehart et al. 2010, Fuentes et al. 2011). Although Chen et al. (2004) did not estimate LC50 values, they observed complete mortality at concentrations as low as 0.75 mg a.e./l, regardless of ph (5.5 vs. 7.5) and food availability. While many of these studies have been conducted at a ph of approximately 8, studies using lower ph conditions have sometimes observed lower toxicity (e.g., ph = 7.5 vs. 6.0; Edginton et al. 2004). Wagner et al. (2013) conducted a meta-analysis of all glyphosate research conducted to date (spanning 37 amphibian species) and found a median LC50 of approximately 2 mg a.e./l with a wide range of variation among amphibian species, amphibian populations, and commercial formulations of glyphosate. Based on definitions from the U.S. Environmental Protection Agency, this range of observed LC50 values indicates that the herbicide is slightly to highly toxic ( pesticide-science-and-assessing-pesticide-risks/tec hnical-overview-ecological-risk-assessment-0). A similar range of LC50 values has been observed in mesocosms with predator cues absent. In the first mesocosm study, Relyea (2005a) observed almost complete death of gray tree frogs, wood frogs, leopard frogs, and American toads when exposed to 3 mg a.e./l of glyphosate. Similar results were described by Jones et al. (2010, 2011). Additional studies have found that the addition of soils in the mesocosms has no mitigating effects on the toxicity of the herbicide for three species of tadpoles (Relyea 2005b). Using just 1 mg a.e./l, Relyea et al. (2005) observed 71% mortality in American toads, 29% mortality in leopard frogs, and no significant mortality of gray tree frogs. Wojtaszek et al. (2004) conducted 2-week exposures in wetland enclosures and found LC50 96-h estimates of mg a.e./l. Additional studies, using green frog and wood frog tadpoles, detected variable results on amphibian survival in wetlands treated with mg a.e./l (Edge et al. 2012, Lanct^ot et al. 2013, Navarro-Martın et al. 2014). Mitigation of herbicide effects on survival by predator cues One of the most striking results of the mesocosm experiment is that the presence of predator cues made the herbicide substantially less lethal to all three species of tadpoles. Under the highest glyphosate concentrations, the addition of predator cues improved survival by 42% in gray tree frogs, 26% in green frogs, and 41% in bullfrogs. These improvements resulted in no effect of the herbicide on two of the three amphibian species when predator cues were present. In contrast, a number of laboratory studies have repeatedly shown that predator cues commonly make glyphosate and other pesticides more lethal to larval amphibians (Relyea and Mills 2001, Relyea 2003, 2004b, 2005c, Giri et al. 2012, Yadav et al. 2013). Moreover, this interaction has been observed in other taxa as well (Qin et al. 2011, Trekels et al. 2011), and it is consistent with synergies between pesticides and competitor stress (Chen et al. 2004, Hanazato and Hirokawa 2004, Jones et al. 2011). The frequent observation that predator cues can make many pesticides more lethal to prey taxa in highly controlled, single-species, laboratory experiments suggests that there are important additional factors present in more natural mesocosm experiments, including thermal stratification. In the only other mesocosm experiment to examine the combined effects of glyphosate and predator cues, Relyea (2012) discovered that predator cues made the herbicide less lethal to an assemblage of spring-breeding amphibians: larval wood frogs, leopard frogs, and American toads. The underlying mechanism was the observation that the water column in outdoor mesocosms experiences temperature stratification which, in turn, can inhibit the mixing of the herbicide (Jones et al. 2010, 2011) for weeks. This results in a higher concentration of herbicide near the surface and a lower concentration near the benthos. This stratification of the herbicide becomes particularly important when predator cues are present because predator cues rapidly scare tadpoles to move down to the benthos. The current study, using an assemblage of summer-breeding amphibian species, clarifies that the phenomenon first observed in Relyea (2012) using an assemblage of spring-breeding amphibians may be quite generalizable across many species of tadpoles. This in no way negates the results of prior laboratory experiments showing that predator cues can make pesticides more lethal, but instead suggests that while predators cues and pesticides can interact 11 November 2018 Volume 9(11) Article e02476

12 synergistically as multiple stressors, the stratification of pesticide concentrations can cause an overwhelming effect that prevents the animals from being exposed to lethal concentrations when behaviorally responding to predator cues. Such stratification can occur in a wide range of water bodies, including lakes, ponds, and wetlands. It also suggests that under windy conditions, which causes the water column (and therefore pesticides) to thoroughly mix, predator cues still possess the potential to make pesticides more lethal. This underscores the context dependence of the outcome, which depends on pesticide concentration, the presence or absence of predator cues, and the presence or absence of astratified water column. The dominant impact of lethal predators When considering the combined impacts of predators and pesticides, a key question is whether the mortality caused by the herbicide is similar in magnitude to the mortality caused by high-risk lethal predators. In other words, while it is clear that the pesticide is highly toxic in the absence of predators, we want to know whether the animals dying from the pesticide would have died at similar or higher numbers if predators were present in the community. For all three amphibian species, the mortality caused by the herbicide alone, which was only significant at the highest concentration, was similar to or less than the mortality caused by the lethal predator, regardless of whether the herbicide was present. At lower herbicide concentrations, the mortality caused by the lethal predator was consistently higher, regardless of whether the herbicide was present. However, there are two important caveats to this conclusion. First, it is important to note that some wetlands (particularly those that dry each summer) can contain few or no predators. In these habitats, the herbicide still has the potential to cause amphibian mortality. Second, the predators used in the current experiment (larval dragonflies) are one of the most lethal invertebrate predators in wetlands. Other invertebrate predators such as notonectids, belostomatids, and dytiscids pose substantially lower risk to tadpoles (Relyea 2001), so the mortality caused by higher concentrations of the herbicide would likely exceed the mortality caused by other species of lethal predators. As noted earlier, few studies have directly compared the impacts of predators and lethal concentrations of pesticides. For example, Boone and Semlitsch (2001) examined the separate and combined effects of lethal predators and lethal concentrations of the insecticide carbaryl on three species of larval amphibians in outdoor mesocosms. Only one of the three species (gray tree frogs) exhibited an effect of predators on survival, but there was no effect of carbaryl on this species. In another example, Hanlon and Relyea (2013) examined the effects of the insecticide on predator prey interactions in laboratory experiment using three species of predators (adult water bugs, Belostoma flumineum; adult newts, Notophthalmus viridescens; and larval dragonflies) and two species of tadpole prey (bullfrogs and green frogs). At the highest endosulfan concentration, there was high (60 100%) mortality of the two tadpole species. In the case of bullfrogs, the mortality was dramatically reduced when any of the predators were combined with the highest endosulfan concentrations. In the case of green frogs, there was not an improvement in survival and the dragonfly predators killed nearly all of the bullfrog tadpoles regardless of endosulfan concentration. Given that we currently do not have many such studies to draw from, it is difficult to draw generalized conclusions regarding the relative mortality rates of pesticides versus lethal predators in aquatic ecosystems. Morphological changes in gray tree frogs Of the three amphibian species in the community, the gray tree frog is the only one that possesses inducible morphology as an adaptive response to predator cues (Relyea 2001). In the current experiment, I found that the gray tree frogs were induced by the caged predators to grow relatively deeper tails and deeper bodies. However, these dimensions were not induced by the herbicide. Instead, the herbicide caused a marginally significant induction of longer and wider bodies. Moreover, the distinct predator induction of red tails in the gray tree frog tadpoles was not significantly induced by the herbicide. This result is in sharp contrast to the morphological induction of wood frog and leopard frog tadpoles caused by the herbicide (Relyea 2012). In that study, both species experienced 12 November 2018 Volume 9(11) Article e02476

13 very similar tail depth induction by caged predators and the herbicide. Moreover, the effects were additive, such that the combination of predator cues and the herbicide caused increases in relative tail depth that were twice the magnitude as either factor alone. A small number of past studies have also found that some pesticides can affect the expression of anti-predator morphology. For example, Hanazato (1991) observed that Daphnia ambigua were inducible by carbaryl to grow spines that are similar to the spines induced by Daphnia predators. Subsequent studies in other species of zooplankton have found that insecticides can either induce anti-predator morphology (Barry 1998, Oda et al. 2011) or inhibit the expression of predator-induced morphology (Barry 1999, 2000, Sakamoto et al. 2006). Collectively, the growing body of research on this topic suggests that pesticide-induced morphological changes may be common in nature and that it may happen because some pesticides are activating similar sets of genes as predator cues. CONCLUSIONS While it is becoming well established that pesticides can have a wide range of unexpected effects on nontarget taxa, these effects can be dramatically modified by the presence of natural stressors including predation and competition. The current study suggests that to know the full impact of these stressors, we need to examine these impacts under more natural ecological conditions. In regard to predatory stress, it suggests that we need to understand the trait-mediated impacts of predator cues as well as the direct lethal effects of predators, since lethal predators can cause high levels of mortality that are of similar magnitude to some pesticides. Future studies should examine the generality of the current study s discoveries. ACKNOWLEDGMENTS My thanks to Nicole Diecks, Jason Hoverman, Lisa Hoverman, Christine Relyea, and Nancy Schoeppner for their assistance with the experiment. I also thank the anonymous reviewers for their thoughtful comments. This study was funded by the U.S. National Science Foundation. LITERATURE CITED Barolo, D Reregistration eligibility decision for glyphosate. EPA 738-R Reregistration Report. U.S. Environmental Protection Agency, Washington, D.C., USA. Barry, M. J Endosulfan-enhanced crest induction in Daphnia longicephala: Evidence for cholinergic innervation of kairomone receptors. Journal of Plankton Research 20: Barry, M. J The effects of a pesticide on inducible phenotypic plasticity in Daphnia. Environmental Pollution 104: Barry, M. J Effects of endosulfan on Chaoborusinduced life-history shifts and morphological defenses in Daphnia pulex. Journal of Plankton Research 22: Berger, G., F. Graef, and H. Pfeffer Glyphosate applications on arable fields considerably coincide with migrating amphibians. Scientific Reports 3:2622. Bernal, M. H., K. R. Solomon, and G. Carrasquilla Toxicity of formulated glyphosate (Glyphos) and Cosmo-Flux to larval Colombian frogs 1. Laboratory acute toxicity. Journal of Toxicology and Environmental Health, Part A 72: Boone, M. D., and R. D. Semlitsch Interactions of an insecticide with larval density and predation in experimental amphibian communities. Conservation Biology 15: Boutin, C., K. E. Freemark, and C. J. Keddy Overview and rationale for developing regulatory guidelines for nontarget plant testing with chemical pesticides. Environmental Toxicology and Chemistry 14: Broomhall, S. D The effects of endosulfan and variable water temperature on survivorship and subsequent vulnerability to predation in Litoria citropa tadpoles. Aquatic Toxicology 61: Broomhall, S. D Egg temperature modifies predator avoidance and the effects of the insecticide endosulfan on tadpoles of an Australian frog. Journal of Applied Ecology 41: Chen, C. Y., K. M. Hathaway, and C. L. Folt Multiple stress effects of Vision herbicide, ph, and food on zooplankton and larval amphibian species from forest wetlands. Environmental Toxicology and Chemistry 23: Clements, W. H., and J. R. Rohr Community responses to contaminants: using basic ecological principles to predict ecotoxicological effects. Environmental Toxicology and Chemistry 28: Dinehart, S. K., L. M. Smith, S. T. McMurry, P. N. Smith, T. A. Anderson, and D. A. Haukos November 2018 Volume 9(11) Article e02476