Methods

General approach: a conservative test of contingency

Our overall approach is to compare the NCE of fish on the abundance of different zooplankton taxa across four experiments performed in different years, but with similar conditions, objectives, measurables and implementation. We are unaware of any prior studies that have examined context dependence of risk effects on prey abundance. The experiments all shared treatments that allow us to examine NCEs of fish on zooplankton abundance by comparing caged-fish and no-fish treatments. The NCE on the abundance of given zooplankton taxa could thus be compared across experiments to identify context dependence of risk effects.
The venue was constructed to capture much of the complexity of natural systems. Each experiment used the same large mesocosms set outdoors (subject to natural environmental variation) and used a natural resource base (phytoplankton) supported by nutrient addition, and a speciose (>10 taxa) zooplankton prey assemblage. Each experiment ran for 3-5 generations of the prey species with the longest generation times (see Peacor et al. 2012), allowing the potential for ecological feedbacks to affect abundance. Although the venue is obviously less complex than a natural system (e.g., lacking in other predator taxa), it incorporates much more of the complexity of a natural system than a typical laboratory experiment that has few prey species, a simple and controlled resource base, and controlled abiotic conditions.
Our approach is thus different than most studies that have identified context dependency in the effect of ecological factors on fitness components over short time scales. Typically, a particular biotic factor, such as prey density, or abiotic factor, such as physical complexity from plants, is systematically varied to investigate the influence of that factor. We ask, rather, could factors that are typically ignored or unmeasured (because they are deemed inconsequential to the study of the risk effect) influence the results of a study of risk effects on abundance in a natural system?
Differences that developed among the complex experimental communities among years are likely caused by factors such as variation in starting abundances that could lead to priority effects (Drake 1991, Chase et al. 2009), small differences in start dates across years, nutrient addition, annual temperature variation, etc. These differences naturally caused variation across experiments, such as the amount of periphyton in the tanks and the absolute and relative abundance of a given taxa. It is these incidental differences that we hypothesize could underlie NCE contingencies. They are small relative to differences among natural systems (e.g. different ponds and lakes), because natural systems have e.g. higher spatial-temporal variation and multiple predators. Subsequently, our test of contingency among experiments is a conservative test for the contingencies relative to those expected in natural systems. Note that by terming the differences “incidental,” we do not mean they were solely due to stochastic environmental differences. Rather, the differences were incidental to the experimental test of NCEs on prey abundance; i.e. they would not be considered important to the outcomes of a study of risk effects of fish on zooplankton abundance.
We provide a description of the experimental methodology that was common to all experiments. Where there were minor differences among experiments, such as small differences in the average size of fish used, a range is provided. Wherever a range is provided in the methods the specific details are provided in a table in the Supporting Information. Other differences among experiments and methodological differences are summarized after the general description and described in more detail in Supporting Information. The timing of manipulations and samplings are given relative to the day experimental treatments were initiated (as provided in Supporting Information), defined as day 0.

Methodology shared among experiments

Outdoor mesocosm experiments were conducted at the E.S. George Reserve (ESGR) of the University of Michigan near Pinckney, Michigan, USA (42°28’N, 84°00’W). Cylindrical cattle watering tanks were employed and contained approximately 1100 L of well water. Tanks were 1.9 m in diameter and 0.75 m tall, and were filled to a depth of 45 cm. Washed sand was added to each mesocosm as a bottom substrate. Each mesocosm was covered with a fiberglass window screen lid to deter colonization by insects. On particularly sunny days, 60% green shade cloth lids were used on top of the window screen lids to reduce heating.
The design of each experiment included the presence/absence of fish kairomone (i.e. chemicals) and additional treatments to explore different questions in each experiment (Supporting Information). A randomized block design was used for each experiment, and replicates ranged from 6-16 depending on the treatment and experiment. The non-consumptive effect of fish was created by maintaining one zooplanktivorous bluegill sunfish, Lepomis macrochirus (standard lengths given in Supporting information) in each of two or three floating cages within each kairomone treatment mesocosom. No-fish treatments consisted of mesocosms with the same number of empty floating cages. Large holes in the sides and bottom of the boxes were covered with fine netting to allow for diffusion of fish kairomones without permitting zooplankton to pass through. In Exps. 1 and 2, three large snails (Planorbella cf. trivolvis >11.2 mm in diameter) were kept inside each cage to graze on periphyton that could grow on the mesh windows. Adding snails was deemed extraneous and not implemented in Exps. 3 and 4.
Fish originated from Patterson Lake, Livingston County, Michigan. In order to ensure fish health and to equalize fish cue (e.g. in case fish differed in cue production) we rotated the fish from the experimental mesocosms to a culture tank once a week. Culture tanks consisted of outdoor mesocosms of ~50 fish of similar size fed zooplankton three times per week. Fish in culture tanks were not fed for 24 hours before being rotated back into the experiment. While in the experimental cages, each fish was fed twice a week, including the day they were added to the cages. To feed fish, an average of 200Daphnia were added per cage. The no-fish cages received an equal amount of Daphnia that were killed by microwaving to ensure a population did not build in the cages.
Calculations of nutrient inputs indicate that nutrients from fish excretion were inconsequential, as they were overwhelmed by other sources, including external supply and internal recycling by zooplankton (See Supporting Information in Peacor et al. 2012). Two experiments also provide evidence that nutrient inputs from fish had no ecological effects. In one, caged fish had no effect on phytoplankton growth in chambers placed outside, but near, the cages and performed soon after (8-10 days) treatments were initiated. In a second in-situ experiment, fish excretion had no effect on phytoplankton growth in chambers placed within the fish cages (Rafalski and Peacor, unpublished).
An initial pulse of nutrients (see below) and a phytoplankton inoculum were added to each tank between 35 and 56 days before the start of the experiment. The phytoplankton inoculum consisted of water collected from ponds in the ESGR and filtered through 35 µm Nitex mesh. A community of zooplankton was added to each tank between day -49 and -27 (Exp. 3 received two inoculums). This initial inoculum came from a single lake (Exps. 2-4) or multiple lakes (Exp. 1, see Supporting Information for details on source lakes). We collected zooplankton using a 64µm zooplankton net, and undesirable animals such as insects (e.g.Chaoborus ) and Hydra were removed prior to adding the inoculum to the mesocosms. To decrease zooplankton heterogeneity among mesocosms, on day -24 to -6 we collected samples of zooplankton from each mesocosm using a 64 µm zooplankton net, mixed all samples, and redelivered subsamples of this mixed community to all mesocosms. The procedure was done twice for Exps. 3 and 4. Although this procedure reduces variation in initial densities among tanks, there will clearly be more variation at treatment initiation than in a more controlled experiment that would start with constant densities.
Inorganic nutrients were supplied to the mesocosms to support phytoplankton growth as a resource for zooplankton at an N:P ratio of 15:1 or 20:1 (Supporting Information). An initial pulse of NH4NO3 and KH2PO4 was added to each mesocosm when phytoplankton were added, and thereafter we supplied maintenance doses weekly. The rate and overall amount of nutrients added were similar among experiments (Supporting Information), though there was some variation in rates and the frequency at which nutrients were added, varying from twice per week to continuous (Supporting Information). In some cases, the weekly rate was reduced to reduce phytoplankton and filamentous algal growth when deemed excessive (Supporting Information). To reduce periphyton growth and to cycle nutrients back to the water column, we collected visible filamentous algae by hand each week, clumped it, allowed it to dry and returned it to the mesocosms. In Exps. 1 and 2, we also added Planorbella cf. trivolvis snails to each mesocosm to reduce periphyton growth.
Zooplankton were sampled on day 31-71. Samples were passed through a 53-64 µm mesh sieve and preserved. Zooplankton were identified and enumerated. Zooplankton identification was resolved to species or genus for Cladocera, adult cyclopoid and adult calanoid for Copepoda (referred to as “cyclopoid” and “calanoid”, exception described below), and ostracods for Ostracoda.
There were a number of methodological differences among experiments due to the specific questions addressed which we summarize here, and describe in more detail in Supporting Information. (1) The zooplankton collection method differed; in Exps. 1 and 2 we subsampled different positions in the tank to determine both zooplankton position in the tanks and overall density, while in Exps. 3 and 4 we measured overall density from combined samples of the entire water column. This difference should not affect our assessment of an NCE of fish on zooplankton abundance. (2) The experiments examined different aspects of risk effects with additional treatments. In three experiments, the additional treatments were included in the no-fish or caged-fish treatment, as they had no influence and their inclusion increased the power to identify treatment effects (Supporting Information). (3) The origin of the zooplankton was the same (a single lake) for Exps. 2-4, but differed for Exp. 1 in which collection was from multiple lakes. (4)Hydra , which are predators of some zooplankton species, were present in Exp. 1, but not Exps. 2-4. (5) Because the size and ecological role of the juvenile stage of cyclopoids can be variable (Burns and Gilbert 1993), we counted only adult cyclopoids. However, in Exp. 4 the counting resources were more limited and thus the cyclopoid counts include both juveniles and adults.
All procedures involving animals were approved by University of Michigan’s Committee on the Use and Care of Animals under approval number 07765 (for experiments 1 and 2) and Michigan State University’s Institutional Animal Care and Use Committee approval no. 02/12-025-00 (for experiments 3 and 4).