Evolution mediates the effects of apex predation on aquatic food webs

Transcription

1 Electronic Supplementary Material Evolution mediates the effects of apex predation on aquatic food webs Mark C. Urban 1 Map of study site Mesocosm experiment methodological details Ecological effects of apex predation Laboratory analyses of eco-evolutionary effects Mesocosm analyses of eco-evolutionary effects

2 1 Map of study site Fig. S1 Study site in Northford, CT. Shapes indicate approximate pond basins. Gray lines indicate 2-m elevation contours. Pond color indicates long-term marbled salamander prevalence (proportion of sample years with a presence recorded) up to

3 2 Mesocosm experiment methodological details I created experimental temporary food webs in 1100-L outdoor mesocosms in the spring immediately following ice-off in natural study ponds. I filled 36 outdoor containers with aged well water in a fenced enclosure covered with 50% shade cloth to simulate natural closed canopy conditions in Storrs, CT. Mesocosms were arranged in three blocks perpendicular to the potential north-south gradient in sunlight. Nine tidbit (Onset Corp., Bourne, MA) temperature loggers distributed across the mesocosm array for the duration of the experiment did not reveal any significant temperature differences along north-south or east-west axes or their interaction (F 3,5 = 0.4, P > 0.4). To each container, I added 25 g of dried deciduous leaves (mostly Quercus spp. and Acer rubrum) and 50 g of pellet-based rabbit food to provide a nutrient base, following standard practice [1,2]. On two occasions in early spring, I inoculated each mesocosm with samples of natural zooplankton and phytoplankton collected in the same six ponds from which experimental spotted salamander populations were collected. I collected phytoplankton by filtering 36 L of water from each of six natural ponds through 750-μm mosquito mesh in the field to exclude large invertebrates. At the same time, I collected 60 1-m zooplankton tow net samples collected from just below the water surface across each pond and added collected zooplankton to the water samples. I then homogenized each sample and added 1-L aliquots of mixed inoculate to each mesocosm from each of the six ponds for a total of 6 L of natural zooplankton and phytoplankton inoculates per mesocosm. 3

4 I collected marbled salamander larvae from five ponds with high population densities at the study site. Each larva was wet-weighed. Initially, larvae were randomly assigned to each mesocosm. Then I adjusted these assignments so that the average mass of predators in each mesocosm (727 g) did not differ significantly among treatments, blocks, or populations (P > 0.4). Four marbled salamander larvae were added to half the mesocosms to create the predation treatment on April 21 st. Four marbled salamander larvae per mesocosm created an initial density of 1.8 larvae per m 2. This value falls within the range of densities estimated in natural ponds over the course of five years at the study site (range = larvae/m 2 ; mean = 0.7 larvae/m 2 ). I used a randomized complete block experimental design. I replicated each of the six spotted salamander populations three times for both the control and predation treatments and placed each replicate randomly in a different north-south spatial block for a total of 36 experimental units. I chose six spotted salamander populations to represent the range in foraging rates found in populations revealed by previous research [3] and confirmed by this research (Figure S4). From April 1-8, I visited each natural population every 2 days in order to collect six newly laid egg masses from each population. Eggs were transported back to the lab and incubated (E-41L2 incubator, Percival Scientific, Inc., Perry, IA) at 8 C (early spring average temperature) with full spectrum lighting and a light-dark cycle that replicated spring field conditions until egg collection was complete. Once egg collection was complete, I divided each egg mass into six 12-egg clusters with a sterile scalpel. After randomly assigning eggs to treatments, egg stage 4

5 did not differ significantly among predation environments (mixed-effects model with predation environment as a fixed effect and population as a random effect: F 1,4 = 0.1, P = 0.803) or among populations (Likelihood Ratio (LR) 1 = 0.1, P = 0.737). I placed eggs into floating mesh rearing containers, which maintained egg masses at the mean depth of eggs in wild ponds (10 cm below surface), in each mesocosm to mimic natural conditions. Four marbled salamander larvae were added to half the mesocosms at this time, which allowed for plastic reactions to develop in response to marbled salamanders prior to hatching. The proportion of hatched larvae midway through the hatching period (May 6 th ) did not differ among treatments, blocks, or populations (P > 0.1). Once most spotted salamander larvae had hatched, I collected 40 surviving larvae from within each mesh rearing container and released them into the corresponding mesocosm on May 8 th. This initial density of 18 spotted salamander hatchlings per square meter is consistent with natural egg densities at the study site (range: eggs/m 2 ; mean = 12.9 eggs/m 2 ) and larval densities more generally [4]. Low survival in one egg rearing container required that I add larvae from a rearing container in another mesocosm from a different block that contained the same population (B-9) and same treatment (no predator). A maximum of ten extra surviving larvae not introduced to the mesocosm were wet-weighted to the nearest 0.1 mg on a Mettler (Columbus, OH) AG204 digital scale to provide an initial mean mass estimate for released larvae. The initial masses of spotted salamander hatchlings did not differ significantly between predation treatments, marbled salamander predation risk, or their 5

6 interaction (P > 0.14), and did not differ significantly among populations (LR 1 = 0.4, P = 0.516). I sampled salamanders and zooplankton communities at five time periods: before spotted salamander hatchlings were released into the mesocosm, and 2, 4, 6, and 10 weeks later. These sampling points correspond to initial conditions unperturbed by spotted salamander feeding (0 weeks), the period of rapid growth into a size refuge from marbled salamander predation and also the stage of maximum spotted salamander predation (2-4 weeks), the low-mortality post-size refuge stage (6 weeks), and metamorphosis (10 weeks). I collected three zooplankton samples using a 15.6-cm diameter vertical pipe sampler [5] positioned in the center, on the edge and midway between the center and edge in each mesocosm to ensure sampling of these potentially different habitats. The total volume sampled was recorded to account for slight variations in mesocosm water depth and sample volume. Zooplankton samples were filtered through 150-um mesh and preserved in 70% ethanol [6]. I estimated zooplankton densities with a Leica MZ 125 stereoscope and identified adult copepods with an Olympus CHBS compound microscope. For each sample, all individuals were identified to the finest taxonomic scale possible using published keys [6-8]. I measured the length of each identifiable taxonomic group up to a maximum of 100 and then used published length-mass regressions [9] to obtain a biomass for each sample. I evaluated changes in spotted salamander survival and zooplankton density and diversity (measured as inverse Simpson s index) and total invertebrate biomass. I analyzed total invertebrate biomass because of the presence of alternative food sources in the 6

7 mesocosms other than zooplankton, especially Dipteran larvae in the families Chironomidae, Culicidae, and Chaoboridae. I separated analyses into those before salamander hatching at week 0 and those afterwards 2-10 weeks. The first set indicates effects of marbled salamanders without the confounded effects of spotted salamander feeding. Spotted salamander evolutionary history was never significantly related to response variables at time 0, which is not surprising given that the salamanders were non-feeding eggs or hatchlings up to this time period and were confined to mesh containers floating on the surface. In these models, I first evaluated the contribution of random spatial effects (East and North) to account for spatial autocorrelation [10] after finding a gradient in zooplankton richness that increased toward the southwest corner of the mesocosm array. East was represented as column number and North was represented as row number. This spatial pattern did not correspond to the insignificant temperature differences among mesocosms (F 3,5 = 0.4, P > 0.4) and thus probably reflected stochastic priority effects in community assembly. Spatial position was not significantly related to zooplankton density, biomass, and Simpson s diversity. In these cases I removed spatial random effects and instead applied the simpler generalized linear model with fixed effects only. In each model, I always included the population of origin of spotted salamanders and mesocosm identity as random effects. This first factor incorporates the relatedness of salamander individuals within a population and is necessary whenever individuals are genetically related (e.g., from the same population) in an experimental design [11]. The second factor incorporates the temporal autocorrelation of responses through time in the same mesocosm. 7

8 3 Ecological effects of apex predation Marbled salamander larvae significantly decreased zooplankton density, biomass and diversity in mesocosms prior to the hatching of spotted salamanders (Figure S2; Table S1; P 0.05). In addition, the marbled salamanders altered the multivariate composition of zooplankton communities as indicated by the significant redundancy analysis. Fig. S2 The effect of marbled salamander larvae on zooplankton density, diversity, and composition before spotted salamander hatching. Red bar indicates mesocosms with four marbled salamanders. For community composition (e), I plot the mean site scores from a redundancy analysis of zooplankton composition based on the presence of marbled salamanders. Error bars indicate SEM; N = 36. 8

9 Table S1 Prey community structure in response to marbled salamander predation in experimental mesocosms Response Effect of predation β (SE) or variance in RDA Random factors included* Test statistic df P-value Density (rank)* (7.26) -- t 34 = Zooplankton Biomass (ln)* (0.39) -- t 34 = < All invertebrate biomass (ln) Simpson s inverse diversity (ln) Diversity (redundancy analysis, RDA) East MCMC upper, lower limits = -2.82, 1.13 < (0.17) -- t 34 = North, East Perm. F = * -- indicates that no random factors were significant and model was analyzed as a generalized linear model. Permutation tests for the mixed-effects model (all invertebrate biomass) and for the RDA do not have degrees of freedom associated with their values. 9

10 Given a significant effect of predation risk on multivariate community composition, I explored individual effects on the proportion of the five most abundant taxonomic groups. Note that I combined Cyclopoid species into a single taxonomic group because 94% were immature and therefore could not be identified to species. Three of the most abundant zooplankton taxa were prevalent throughout the experiment; however, two occurred at negligible densities early on but then grew highly abundant by the end of the experiment. As a result, I analyzed the effect of marbled salamanders on early taxa (S. mucronata, D. ambigua, and Cyclopoid copepods) at the first sampling point but analyzed two late-establishing taxa (Chydorus sphaericus and Ceriodaphnia dubia) at the last sampling point. The interpretation of late patterns is more complicated than for early patterns because late samples combine the direct effect of marbled salamander predation with the indirect numerical and behavioral effects of marbled salamanders on spotted salamander predation. Marbled salamanders were associated with a significant increase in S. mucronata, D. ambigua and C. sphaericus and a significant decrease in Cyclopoid copepods (Figure S3; Table S1; n = 36; P = 0.022, P = 0.005, P = 0.040, P = 0.005). C. dubia was not significantly affected by marbled salamander predation (P = 0.843). 10

11 Fig. S3 The effect of marbled salamander predation on the five most numerically dominant zooplankton taxa in mesocosms. Red bar indicates mesocosms with marbled salamanders. Proportions are indicated at the first sampling period for a c and at the last sampling period for d and e, where these latter taxa were at low numbers initially. Error bars indicate SEM. Note different y-axis scales among subpanels. N = 36 for all panels. 11

12 Table S2 Changes in the proportions of dominant zooplankton taxa in response to marbled salamander predation in experimental mesocosms Response Effect of predation β (SE) Random factors included* Test statistic df P-value Scapholeberis mucronata Cyclopoid copepods (0.77) Early response (no spotted salamanders) 3.06 (1.28) -- t 34 = North, East, overdispersion z 31 = Daphnia ambigua 2.96 (0.99) -- t 34 = Late response for species with low numbers initially Chydorus sphaericus 2.01 (0.94) -- t 34 = Ceriodaphnia dubia (3.14) overdispersion z 34 = * -- indicates that no random factors were significant and model was analyzed as generalized linear model. I examined how marbled salamander predation affected the proportion of these species in the last, rather than the first, sample because these species were at near zero numbers on the first sampling date but became highly abundant by the end of the experiment. 12

13 4 Laboratory analyses of eco-evolutionary effects Fig. S4 Effect of evolutionary history of predation risk on spotted salamander preference for zooplankton taxa in laboratory experiments. Each symbol indicates a separate spotted salamander population arranged along an axis of increasing predation risk. An asterisk next to a taxon s picture indicates a significant difference from zero for that taxon across all populations. If a regression line is present, it indicates a significant effect of evolutionary history. Error bars indicate SEM; N =80. 13

14 5 Mesocosm analyses of eco-evolutionary effects Table S3 Prey biomass through time (2-10 weeks) Factors Fixed effects Estimate (SE) MCMC lower 95th MCMC upper 95th P Predator presence (PP) (0.495) Predation history (PH) (1.573) Time (T) (0.421) Time 2 (T 2 ) (0.034) PP x T (0.067) PH x T (0.573) PH x T (0.046) Random effects Var (SD) LR df P Population (0.160) Individual (0.638) North (0.997) < East (0.803)

15 Table S4 Zooplankton density through time (2-10 weeks)* Factors MCMC MCMC Fixed effects Estimate (SE) lower upper P 95th 95th Predator presence (PP) (12.36) Predation history (PH) (16.88) Time (T) (4.66) < Time 2 (T 2 ) 1.49 (0.37) < PP x T 7.02 (1.75) < Random effects: Var (SD) LR df P Population Individual (0.200) (0.555) North (0.851) < East (0.640) * Ranked data used after a transformation to normality could not be found. 15

16 Table S5 Inverse Simpson s diversity through time (2-10 weeks) Factors Estimate (SE) MCMC lower 95th MCMC upper 95th P Fixed effects: Predator presence (PP) (0.160) Predation history (PH) (0.155) Time (T) (0.045) < Time 2 (T 2 ) (0.004) < PP x PH (0.219) Random effects: Var (SD) LR df Population 1.28e-17 (3.58e-9) Individual (0.130)

17 Fig S5 (next page) Changes in the proportion of A) Cylopoid copepods, B) C. sphaericus, and C) S. mucronata through time with and without marbled salamander predation (top) and depending on spotted salamander evolutionary history of predation (bottom), averaged across all time periods (no significant predation history x time interaction). In C, there was a significant predation history by time 2 interaction, which is displayed in the bottom subpanel. For presentation purposes, linear regression lines are plotted in the bottom graph for time periods where R 2 values > 0.2. In all graphs, means of fitted values from mixed-effect models are depicted ± SEM; N =

18 A B C 18

19 Table S6A Change in the proportion of Cyclopoid copepods through time (2-10 weeks) Factors Fixed effects: Estimate (SE) z P Predator presence (PP) (0.56) Predation history (PH) 0.74 (0.33) Time (T) (0.43) < Time 2 (T 2 ) 0.14 (0.03) 4.16 < Random effects: Var (SD) LR P Population 0.10 (0.31) Mesocosm 0.01(0.07) Individual* 6.63 (2.57) < North 1.13 (1.06) < East 2.43 (1.56) *I modeled overdispersion directly by including a random effect for each unique mesocosmtime combination as per the suggestion of [12]. 19

20 Table S6B Change in the proportion of Chydorus sphaericus through time (2-10 weeks)* Factors Fixed effects: Estimate (SE) z P Predator presence (PP) 4.92 (2.04) Predation history (PH) 1.73 (2.11) Time (T) 3.00 (0.34) 8.91 < PP x PH (2.93) Random effects: Var (SD) LR P Population 0.12 (0.34) Mesocosm 0.86 (0.93) Individual 3.20 (1.79) 460 < North 0.38 (0.62) East 0.08 (0.29) * This significant effect was conditioned on removing one data point from the last time period from a mesocosm in which spotted salamanders had become extirpated. In this mesocosm, more than 80% of the zooplankton were C. sphaericus (unlike the mean of 23%), and this data point had high enough leverage according to ref. [13] to suggest removal provided a justified external reason (leverage = 0.25, mean leverage of dataset = 0.11). The external reason in this case is because this was the only mesocosm in which all the spotted salamanders died. Without this point removed, the effect of predator presence x predation history is no longer significant (P = 0.149). Therefore the significant fixed effects results should be treated with caution. I modeled overdispersion directly by including a random effect for each unique mesocosmtime combination as per the suggestion of [12]. 20

21 Table S6C Change in the proportion of Scapholeberis mucronata through time (2-10 weeks)* Factors Fixed effects: Estimate (SE) z P Predator presence (PP) 4.75 (1.45) Predation history (PH) (1.02) Time (T) 4.45 (0.40) < Time 2 (T 2 ) (0.03) < PP x T (0.55) < PP x T (0.05) 5.19 < PH x T (0.01) Random effects: Var (SD) LR P Population 0.11 (0.34) 142 < Mesocosm 0.67(0.82) 174 < Individual* 2.72 (1.65) < North 0.12 (0.34) 152 < East 0.11 (0.34) < * I modeled overdispersion directly by including a random effect for each unique mesocosmtime combination as per the suggestion of [12]. 21

22 References 1 Skelly, D. K Experimental venue and estimation of interaction strength. Ecology 83, Relyea, R. A Local population differences in phenotypic plasticity: predator-induced changes in wood frog tadpoles. Ecol. Monogr. 72, Urban, M. C Risky prey behavior evolves in risky habitats. Proc. Natl. Acad. Sci. USA 104, Brodman, R Effects of intraguild interactions on fitness and microhabitat use of larval Ambystoma salamanders. Copeia 1996, Paggi, J. C., Mendoza, R. O., Debonis, C. J. & Jose de Paggi, S. B A simple and inexpensive trap-tube sampler for zooplankton collection in shallow waters. Hydrobiologia 464, Williamson, C. E. & Reid, J. W Copepoda. In Ecology and classification of North American freshwater invertebrates (ed. J. H. Thorp & A. P. Covich), pp Boston: Academic Press. 7 Haney, J. F An Image-based Key to the Zooplankton of the Northeast, USA Version 4.0 released 2010: University of New Hampshire Center for Freshwater Biology. 8 Dodson, S. I. & Frey, D. G Cladocera and other Branchipoda. In Ecology and classification of North American freshwater invertebrates (ed. J. H. Thorp & A. P. Covich), pp Boston: Academic Press. 9 McCauley, D. E The estimation of the abundance and biomass of zooplankton in samples. In A manual on methods for assessment of secondary productivity in fresh waters (ed. J. A. Downing & F. H. Rigler), pp Oxford: Blackwell Scientific Publications. 10 Crawley, M. J The R Book. Chichester: John Wiley & Sons. 11 Bolker, B. M., Brooks, M. E., Clark, C. J., Geange, S. W., Poulsen, J. R., Stevens, M. H. H. & White, J.-S. S Generalized linear mixed models: a practical guide for ecology and evolution. Trends Ecol. Evol. 24, Warton, D. I. & Hui, F. K. C The arcsine is asinine: the analysis of proportions in ecology. Ecology 92, Neter, J., Kutner, M. H., Nachtsheim, C. J. & Wasserman, W Applied linear statistical models. Boston: McGraw-Hill. 22