Two short-term mesocosm experiments (PU1 and PU2, Figure 1) with in situ plankton communities from the Peruvian coastal upwelling region were monitored in 12 shipboard mesocosms during the M77/3 cruise off Peru (Franz et al., 2012b, 2013a,b; Hauss et al., 2012). The objectives of the PU1 and PU2 mesocosm studies were to identify the influence of inorganic nutrient concentrations and ratios on the development of plankton biomass and community composition across trophic levels in the Peruvian Upwelling region (Franz et al., 2012a,b, 2013a,b; Hauss et al., 2012). PU1 consisted of three nutrient treatments: An unenriched control with ambient nutrient concentrations (N:P = 3.4), an enrichment with high (N:P = 20) and an enrichment with low (N:P = 2.8) dissolved inorganic N:P ratios (Figure 1). PU2 had four nutrient treatments: Two high-N:P treatments (N:P = 16 and 8) and two low-N:P treatments (N:P = 5 in the unamended control and 2.5) (Figure 1). All mesocosms were covered with a shading net to achieve ≈30% of the ambient light intensity (Figure 1). The initial water samples obtained from Niskin bottles mounted on a CTD were filtered (pre-screened) through a 200 μm mesh to remove mesozooplankton from all mesocosms of PU2 and from two mesocosms per treatment of PU1. As in Hauss et al. (2012), we do not distinguish between mesocosms with and without mesozooplankton. However, the microzooplankton community was dominated by dinoflagellates in PU1 and by ciliates in PU2. All mesocosms were restocked with 5 μm-filtered ambient surface seawater on days 3 and 5 of the experiments, due to the large amounts of water required for sampling (Figure 1; Franz et al., 2012a,b, 2013a,b). Trace metal and silicate compounds were added to avoid trace metal and silicate limitation at the start of each experiment, and also on day 5 in PU2 only.

Our optimality-based food-chain model defines up to three trophic levels, representing dissolved inorganic nutrients (NN), phytoplankton (P), and one or two zooplankton compartments (Z) (Figure 2 and Figure S1). The phytoplankton compartment is represented by 4 state variables allowing for dynamic C:N:P:Chlorophyll (Chl) ratios (see Appendix in Supplementary Material, Equations 3–10), whereas the zooplankton compartments have constant C:N:P ratios (see Appendix in Supplementary Material, Equations 11–13, and Table 1).

For the phytoplankton compartment, we employ the OCM (Pahlow et al., 2008, 2013; Pahlow and Oschlies, 2013). In the OCM, the phytoplankton phosphorus quota (QPP, defined as the P:C ratio) is limiting nitrogen assimilation and the nitrogen quota (QPN, the N:C ratio) controls nutrient uptake and carbon-fixation. Thus, both N and P always colimit growth in the OCM. The OCM explicitly represents light and dark respiration by light-dependent and light-independent respiration terms. For simplicity, we do not simulate a diurnal light cycle, but multiply daytime photosynthesis and light-dependent (but not dark) respiration with the day-length (0.5).

The OCM is coupled to the optimal current-feeding model for zooplankton (OCF, Pahlow and Prowe, 2010). The OCF is built on trade-offs among foraging activity, assimilation efficiency and respiration. We employ unaltered pre-calibrated parameter sets by Pahlow and Prowe (2010) to represent ciliate and dinoflagellate behavior. The only exception is the prey capture coefficient (ϕ), which is reduced for non-preferred prey in order to mimic preferential feeding (see below). We assume temporally constant (homeostatic) microzooplankton elemental stoichiometry. Thus, the excess C, N, or P, which cannot be assimilated, is excreted in dissolved form (Kiørboe et al., 1996). To reduce model complexity we do not differentiate between excretion and egestion of particulate matter. The excretion terms for C, N, and P are given by the difference between ingestion and assimilation, corresponding to the difference between the variable elemental C:N:P ratio of the prey and the predefined constant elemental C:N:P ratio of the microzooplankton compartments (see Appendix in Supplementary Material, Equations 11–13, and Table 1).

We use observations from the PU1 and PU2 shipboard mesocosm experiments of the M77/3 cruise (Franz et al., 2012b, 2013a,b; Hauss et al., 2012) to determine the initial conditions for the model-setup and to assess model performance for the duration of the experiments. Dissolved inorganic nitrogen and phosphorus (DIN and DIP, respectively) represent all dissolved nitrogen and phosphorus compounds available to phytoplankton. For simplicity we do not address the dissolved organic matter (DOM) pool, since there are no clear trends in DOM concentrations throughout the experiments (Franz et al., 2012b, 2013b). Initial phytoplankton C, N, P are calculated from (averaged) observed POC, PON, POP concentrations (Franz et al., 2012b, 2013a,b; Hauss et al., 2012), from which we subtract the (averaged) observed dinoflagellate, ciliate, and bacterial biomasses multiplied with assumed zooplankton and bacterial N:C and P:C ratios, QZN, QBN, QZP, and QBP, respectively. Assumed QZN and QZP are given in Table 1. We apply the same N:C and P:C ratios to bacteria (Chrzanowski and Grover, 2008; Pahlow et al., 2008; Zimmerman et al., 2014), which are only used for the calculation of the initial phytoplankton C, N, P in this study. Thus, our initial phytoplankton PON and POP concentrations vary slightly between the different simulations of the same mesocosms, depending on the assumed zooplankton and bacterial N:C and P:C ratios.

We initialize our model with observations for the first day (day 0) for PU1 and data for the second day (day 1) for PU2, due to the lack of initial POC, PON, and POP measurements for PU2. We account for initial (day 1) differences between individual mesocosms within each treatment of PU2 (Hauss et al., 2012) with three ensemble simulations for each treatment. Our PU1 and PU2 model simulations are all run for 7 days.

We simulate the restocking of the mesocosms of both experiments by adding DIN and DIP, according to the concentrations and mixing ratios of the restocking medium indicated in Figure 1 on days 3 and 5 of both experiments (Franz et al., 2012b, 2013a,b; Hauss et al., 2012). All remaining model compartments are multiplied with dilution factors, i.e., the ratio of the actual mesocosm water volume before restocking over the total (restocked) mesocosm water volume (f_dil = A_V:I_V). We assume that the restocking medium (5 μm-filtered ambient surface seawater) contained only water and inorganic nutrients, since no zooplankton or phytoplankton counts were performed.

Different assumptions represented by the model structure about the trophic interactions and stoichiometry might result in different interpretations of the observations of both mesocosm experiments. To test our assumptions we therefore set up several model configurations (Figure 2) to simulate conceptually possible food web interactions in the two mesocosm experiments. The different model configurations differ in model complexity in terms of the number of trophic levels resolved and/or the trophic strategies of the microzooplankton community, as well as zooplankton stoichiometry.

At first we apply the (OCM) for phytoplankton (Pahlow et al., 2013) and two nutrients (NN), DIN and DIP. This model configuration with only one trophic level helps us to find out whether bottom-up control alone could explain the development of nutrients and phytoplankton.

We investigate the effects of top-down control with the (OCF) for zooplankton (Pahlow and Prowe, 2010) where we progressively increase the number of trophic levels by representing nutrients, (NN), phytoplankton (P) and up to two microzooplankton types, Z1 and Z2 (Figure 2, Figure S1). Since the microzooplankton community in the mesocosms was identified as comprising ciliate and dinoflagellate species (Hauss et al., 2012), the foraging strategies in our model are defined by the dinoflagellate and ciliate parameter sets (Pahlow and Prowe, 2010; Table 1 and Table S1). In all simulations of each of the different model configurations we use the same pre-calibrated parameter set for all treatments and vary only the initial conditions of our state variables (see Appendix in Supplementary Material, Equations 1–7), according to the corresponding mesocosm observations as described above. In this study we analyse both specialist (strictly herbivorous or carnivorous) and omnivorous feeding of microzooplankton (Figure 2 and Figure S2), to find out whether patterns of food preferences or foraging strategies could explain the observed differences in the two mesocosm experiments between and within the mesocosm treatments.

Furthermore, we consider stoichiometric plasticity of the microzooplankton community as a possible physiological response to changes in food quality. Therefore, we imitate the N and P requirements of higher trophic levels by applying a wide range of elemental microzooplankton N and P quotas (N:C and P:C ratios, respectively). The suffix “-zooQP” in the configuration name indicates that we applied a higher microzooplankton P quota.

These different model configurations are designed (1) to determine the minimum trophic levels of the model structure required to match observed patterns in the mesocosms, and (2) to investigate the effects of nutrient stoichiometry, related to the observed niche substitution of microzooplankton between PU1 and PU2.

The simplest (NNP) configuration contains only the nutrient (NN) and phytoplankton (P) compartments and has 6 state variables (see Appendix in Supplementary Material, Equations 1–6 and Figure 2). The intermediate (NNPZ) configuration contains a second trophic level (one additional state variable), the zooplankton guild (Equation 7 and Figure 2). Additional information on the sensitivity configurations with dinoflagellates, specialists and omnivores, and the three trophic level configurations can be found in the electronic supplement (e.g., Figure S1).

In the NNP configuration, primary production of the phytoplankton compartment is the only process responsible for “bottom up” control. The NNP configuration lacks phytoplankton mortality, because we do not employ a zooplankton grazing function representing top-down control. We modify the phytoplankton parameters within the ranges given by Pahlow et al. (2013) and include dynamic photo-acclimation to match the onset of the phytoplankton bloom in the mesocosms during the first 3 days. We employ faster Chl dynamics (see Appendix in Supplementary Material, Equations 8–10) than in Pahlow (2005), which compares better with the observed initial time-course of Chl and the Chl:C ratio in the mesocosms.

We simulate top-down control in the specialist, strictly herbivorous, NNPZ-s configuration by microzooplankton grazing only on phytoplankton. In the omnivore NNPZ-o configurations we also allow top-down control, hereafter called intraguild predation, within the microzooplankton compartment (Figure 2). Intraguild predation is seen in our model as controphic species predation rather than cannibalism, since we assume that each microzooplankton compartment represents many species encompassing a range of sizes (Stav et al., 2005). We differentiate between the NNPZ-s (specialist) and NNPZ-o (omnivore) configurations by means of different microzooplankton feeding behavior represented by different prey capture coefficients (ϕ) to simulate variations in food preferences. The preferred food source is associated with the highest ϕ, i.e., the ϕ according to Pahlow and Prowe (2010) (Figure 2). We apply lower prey capture coefficients for predation within the microzooplankton guild. Owing to a lack of observations, we pragmatically set ϕ for intraguild predation to one-half of the ϕ for the next lower trophic level. In this way, we implicitly split each zooplankton compartment into equal contributions of intraguild predators and prey.

We focus here on three configurations, the NNP and the two omnivore NNPZ-o with a low P quota and NNPZ-o-zooQP with a higher P quota (see below). Please consult the electronic supplement for the description and set-up of the specialist (herbivore) NNPZ-s and NNPZ-s-zooQP configurations (Figures S2, S3).

Bottom-up processes appear to have dominated ecosystem dynamics during the first 3 days of the mesocosm experiments, providing constraints for our phytoplankton parameters. For the NNP configuration, it proved impossible to match the first 3 days of the mesocosm behavior without dramatically overestimating phytoplankton biomass toward the end of PU1 and PU2, when the model mesocosms entered the stationary phase (Figures 3, 4, respectively). Most likely this results from the lack of top-down control (grazing mortality) of phytoplankton. Thus, predation losses and nutrient remineralization had a significant impact on the development of the mesocosm ecosystems. Surprisingly, phytoplankton N:P nevertheless matches the observations quite well. Moreover, observed phytoplankton N:P variations of both PU experiments between treatments are rather minor compared to variations within treatments (Figures 3C, 4C).

The specialist (herbivore) model configuration (NNPZ-s) represents the simplest food-web structure to include top-down control (Figure 2). In the NNPZ-s configuration, phytoplankton declines too rapidly and nutrients rise too high toward the end of the experiments in all PU1 (Figures S2A1–A3,B1–B3) and the low-N:P (DIN:DIP < 6) treatments of PU2 (Figures S3A3,A4,B3,B4). Phytoplankton declines to sufficiently low concentrations to cause food limitation in the microzooplankton compartment and microzooplankton biomass is overestimated toward the end of the low-N:P treatments of PU2 (Figures S2B2,B3). Microzooplankton is not food limited in the other simulations (ingestion saturation ≈ 1) (Figures S2B3,B4). Although the microzooplankton biomass in the PU1 experiment was dominated by dinoflagellates, ciliate parameters according to Pahlow and Prowe (2010) give the best fit of the model to the data in both experiments (Table 1, Figures 5, 6 and Figures S2, S3).

The omnivore (NNPZ-o) configuration yields a fair reproduction of the phytoplankton biomass (Figures 5B1, 6B) and also matches microzooplankton biomass in all PU1 simulations (Figure 5B) and in those for the high-N:P (DIN:DIP > 6) treatments of PU2 (Figures 6B1,B2). Compared with the NNPZ-s configuration (Figures S2, S3), model phytoplankton represents the observations better for the high-N:P treatments of PU1 (Figure 5B1), while it agrees better for the low-N:P treatments for PU2 (Figures 6B3,B4). Remineralization and microzooplankton biomass are overestimated in the low-N:P treatments of PU2 (Figures 6A3,A4,B3,B4). The NNPZ-o simulation largely reproduces the observations for inorganic nutrients, and phytoplankton and zooplankton biomass in the high-N:P treatments in both experiments (Figures 5A1–B1, 6A1,A2,B1,B2). The model overestimates the Chl:C ratios for PU1 but those for PU2 appear to be captured quite well (Figures 5C, 6C). In summary, the NNPZ-o configuration appears capable of reproducing the high-N:P but not the low-N:P treatments of both experiments (Figure 7).

We analyse the sensitivity of our simulations with respect to variations in the parameters describing N and P subsistence quotas for phytoplankton and zooplankton N:C and P:C ratios in an attempt to unravel the causes of the poor fit of the NNPZ-o simulations to the low N:P treatments. These parameters determine the stoichiometry of our plankton compartments. No significant improvement in model performance is achieved by varying the phytoplankton subsistence quotas or the microzooplankton N:C ratio. However, when we apply a higher microzooplankton P:C ratio (QZP set to 0.0195 molP molC⁻¹, Table 1) in the NNPZ-o-zooQP configuration, which is otherwise the same as the NNPZ-o configuration, we obtain the best results for the low-N:P treatments for both PU1 and PU2. Nevertheless, the NNPZ-o-zooQP configuration fails to reproduce the high-N:P treatments (Figures 5–7). Thus, a high QZP works for low-N:P but not high-N:P treatments. We also conduct simulations with more complex model configurations, where we employ two zooplankton compartments to simulate the ciliate and dinoflagellate communities separately (NNPZZ configuration, Figure S1). However, these do not perform better than the NNPZ-o and NNPZ-o-zooQP configurations.

Considering all observations, model configurations, and processes together, the NNPZ-o configuration with the low microzooplankton P:C quota best reproduces the high-N:P treatments, whereas the NNPZ-o-zooQP best reproduces the low-N:P treatments in both experiments (Figure 7).

The NNP configuration does not describe phytoplankton mortality. Addition of the microzooplankton compartment in NNPZ-s and NNPZ-o introduces top-down control and thus balances phytoplankton growth (bottom-up control). The suppression of phytoplankton and overestimation of remineralization in the specialist (NNPZ-s) simulations of the low-N:P (DIN:DIP < 6) treatments prompted us to investigate further possible top-down controls within the microzooplankton community. Intraguild predation in the omnivore (NNPZ-o) configuration does indeed control microzooplankton growth. We conclude that at least two trophic levels and omnivory are needed in our model to reproduce the observed behavior of the mesocosm plankton communities.

For both PU1 and PU2 it proves impossible to reproduce with just one microzooplankton P:C ratio (QZP the high- and low-N:P treatments at the same time (Figure 7 and Appendix in Supplementary Material, Equations 14, 15). The high-N:P treatments agree better with a lower microzooplankton QZP (NNPZ-o configuration), whereas the low-N:P treatments agree better with a higher microzooplankton QZP (NNPZ-o-zooQP configuration) (Figure 7). Thus, we hypothesize a flexible elemental composition of the microzooplankton.

Flexible microzooplankton stoichiometry might compensate partly for low food quality in terms of C:N:P composition of the prey. At first sight, our results seem to indicate a relationship between external nutrient stoichiometry and microzooplankton internal elemental composition. If the initial inorganic DIN:DIP ratio is high, microzooplankton with a lower P:C ratio likely grow better, although phytoplankton N:P and P:C do not differ strongly (Figure 8). In all (NNPZ-o and NNPZ-o-zooQP) simulations for both PU1 and PU2, phytoplankton N:C showed a much clearer relation than N:P or P:C to the low- and high-N:P treatments (Figure 8): In the first 2 days of our model simulations, phytoplankton N:C ratios develop in groups according to the initial DIN:DIP ratio (Figure 8), implying a distinction in phytoplankton food quality in terms of N:C, rather than N:P, between the high- and low-N:P treatments. While the observed phytoplankton N:C ratios shown in Figure 8 do not reveal such a clear separation according to the initial DIN:DIP ratio, they mostly agree with our model simulations, except during the second half of the PU2 experiments (Figure 8). Neglecting detritus, which was not quantified in the original mesocosm experiments, might have caused some of this discrepancy. Part of this discrepancy might also be due to the fact that in our model the microzooplankton C:N:P stoichiometry was temporally constant, and thus cannot adjust in response to changes in phytoplankton food quality during the model simulations.

In Meunier et al. (2012b) selection experiments, the microzooplankton P:C ratio was also higher when the phytoplankton N:C ratio was lower and vice versa (their Table 4). We thus hypothesize that microzooplankton adjust their internal P:C ratio in response to food quality in terms of prey N:C ratio. Thus, differences in phytoplankton N:C ratios during the first days of the experiments could have served as the signal to which the microzooplankton P:C ratio responded (Figure 8C).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.