Introduction

How organisms take up resources, grow, reproduce, or interact with competitors, pathogens, or consumers can be strongly affected by the presence of other species in an ecosystem (Stachowicz 2001; Olff et al. 2009; Kéfi et al. 2012). Interactions among species can affect the functioning of ecosystems by regulating fluxes of energy and matter, ecosystem productivity and metabolism, or by mediating the response of ecosystems to perturbation (Loreau et al. 2001; Harmon et al. 2009; Chapin et al. 2011). Some species interactions are more important than others for ecosystem functioning (Angeliniet al. 2011; Falkenberg et al. 2012), such that modifying these interactions can have disproportionate impacts on ecosystems: so called foundation species (Dayton 1972) can define much of the structure of a community by creating locally stable conditions for other species. Disturbances can influence how foundation species can individually or interactively affect multiple ecosystem components (Ellisonet al. 2005; Darling & Côté 2008), potentially causing surprising effects on ecosystems (Paine et al. 1998). Such complexity regarding the interplay between interactions of important species and environmental change makes forecasting ecosystem responses to increasing anthropogenic disturbances particularly challenging (Petcheyet al. 2015).
Eutrophication is a threat to aquatic ecosystems worldwide (Smith et al.1999; Smith 2003), and there is growing evidence that nutrient loading can cause both gradual and sudden shifts in ecosystem state, depending on the nature and strength of species interactions (Carpenter 2005). The presence or absence of different key species, including macrophytes, benthic and pelagic grazers, and phytoplankton, are thought to define ecosystem responses to nutrient perturbations through positive and negative interactions among them (Scheffer et al. 1993; Kéfi et al. 2016). In the network of species interactions in shallow lakes (Scheffer et al.1993), for example, a key interaction is the competition between macrophytes with phytoplankton communities for dissolved nutrients and light (Schefferet al. 1993; Ibelings et al. 2007). Assemblages of macrophytes, that are considered to be important foundation species (Scheffer et al. 2003; Kéfi et al. 2016), are competitively dominant at low nutrient loading, and can persist at intermediate nutrient loading via a positive feedback between macrophyte growth and water transparency (Carpenter & Lodge 1986; Jeppesen et al. 1998). Compared to macrophytes, fewer models have investigated how benthic grazers can influence the dynamics of ecosystem responses to nutrient pulses. However, the musselDreissena polymorpha can directly consume large amounts of phytoplankton, (Johengen et al. 1995; James et al. 1997), and its occurrence has coincided with dramatic changes in water clarity of some lake ecosystems (Ibelings et al.2007).
Beside their expected negative effect on phytoplankton biomass, foundation species like macrophytes and grazers can also affect other ecosystem properties such as dissolved organic matter (DOM) and oxygen metabolism (Schefferet al. 1993; Olff et al. 2009; Kéfi et al. 2016). As such, they can mediate how external disturbances reverberate through the network of biological and abiotic interactions in aquatic ecosystems (Narwani et al. 2019). Such effects can culminate in changes in both the mean and variance of ecosystem parameters, which can sometimes foreshadow a sudden shifts in ecosystem state (Carpenteret al. 2011; Scheffer et al. 2012; Gsell et al.2016). Studies using high resolution measurements are particularly useful for tracking the mean and variance of ecosystem metrics, for example, phytoplankton biomass and rates of ecosystem metabolism such as net primary productivity and respiration (Carpenteret al. 2011; Batt et al. 2013; Nielsen et al.2013). These processes are largely driven by the autotrophic lake community, both benthic (i.e. macrophytes) and pelagic (i.e. phytoplankton), but can also be affected by DOM dynamics associated with the growth and decay of biomass (Catalán et al.2014). Photosynthesis and respiration rates can be modeled with relatively high precision using repeated measurements of dissolved oxygen and water temperature (Staehr et al.2010), and have been used to assess ecosystem resistance and resilience (Batt et al. 2013). However, such approaches, using an array of multiple high-resolution sensors (16 sondes with 5 parameters/sonde), have never been applied in factorial manipulations of foundation species in aquatic ecosystems.
When facing disturbance, interactions between foundation species can cause non-additive effects on ecosystem dynamics that are difficult to anticipate (Narwani et al. 2019), and this may impair our ability to quantify resistance and resilience of ecosystems with a particular species configuration (Schefferet al. 1993; Allgeier et al. 2011; Kéfi et al.2016; Thompson et al. 2018). However, only very few studies have attempted to experimentally disentangle singular and synergistic effects of key species on ecosystem dynamics in response to changing environmental conditions (Stachowicz 2001; Angelini et al. 2011; Falkenberg et al. 2012). In shallow lake ecosystems, the presence of either macrophytes and mussels have been linked to increased capacity to maintain a clear water state with low phytoplankton abundances (Jeppesenet al. 1998; Bierman et al. 2005; Ibelings et al.2007). Current theory suggests that both species may facilitate the presence of each other: macrophytes can provide habitat forDreissena mussels to settle on (Ibelings et al. 2007; Karatayev et al. 2014b), and mussels can actively decrease local turbidity, thus improving environmental conditions for submerged macrophytes (Ibelings et al.2007). Such facilitation is a common phenomenon in ecological communities (Stachowicz 2001; Angelini et al. 2011; Falkenberg et al. 2012), especially for foundation species, like macrophytes and mussels. However, there is also potential for antagonistic interactions between macrophytes and mussels that could unfold under nutrient perturbation scenarios: polyphenols and fatty acids produced by macrophytes to inhibit phytoplankton growth (Korner & Nicklisch 2002; Hilt & Gross 2008) may be harmful to filter feeding organisms (e.g. mussels), whereas mussels can shift the composition of phytoplankton communities to species that might be less affected by allelochemicals (Vanderploeg et al. 2001; Fishman et al. 2010).
Here, we monitored freshwater ponds in high resolution to experimentally test how a disturbance scenario characterized by multiple nutrient pulses over two years affects pond ecosystem dynamics with and without two important foundation species. We manipulated the presence and absence of the macrophyte Myriophyllum spicatum and the musselDreissena polymorpha , two important foundation species that are common in freshwater ecosystems worldwide. In a factorial pond experiment, we perturbed all ecosystems by progressively increasing the input of inorganic nutrients and quantified the dynamics of several biotic and abiotic ecosystem parameters. The goal was to investigate how the presence and absence of two important foundation species affects the dynamics of a suite of ecosystem parameters during the process of pond eutrophication (our disturbance regime). Specifically, we aimed at characterizing how the nature of interactions between the two species (additive vs. non-additive, Figure 1A) was affected by nutrient perturbation. Contrary to our expectation of more stable clear water conditions under the presence of both foundation species, after both periods of nutrient perturbation (year 1, and 2), we found strong non-additive, antagonistic effects in several ecosystem parameters (see also Narwani et al. 2019, for the results from the first year of manipulation). Our results demonstrate how interactions between key species can drastically change under disturbance regimes, emphasizing the importance of understanding how species interaction networks, and how they change over time, can affect ecosystem responses to disturbance (Schefferet al. 1993; Ibelings et al. 2007; Kéfi et al.2016).