Discussion

Perturbation of the pond ecosystems with nutrients evoked strong responses in all ponds, which were dependent on the presence of foundation species and, in some cases, their co-occurrence. As expected, all nutrient pulses led to strong increases in phytoplankton abundances across all treatment combination, which, at first, was mediated by the presence of either Myriophyllum or mussels in the single species treatment. However, when both Myriophyllum and Dreissenawere present within a pond, nutrient additions led to a contrasting pattern: phytoplankton biomass in these ponds increased stronger than in the presence of a single species or when none of the two species were present. These patterns suggest strong non-additive interactions between macrophytes and mussels that affected phytoplankton biomass during and following the disturbance periods.
Mediation of phytoplankton blooms under increased nutrient loading by either macrophytes or mussels alone was expected, and is in agreement with a large body of previous theoretical and empirical work (van Neset al. 2007; Iacarella et al. 2018; Yamamichi et al. 2018). Macrophytes can keep phytoplankton biomass in the water column at lower levels compared to ecosystems that lack macrophytes. Such control of phytoplankton biomass by macrophytes is often linked to their competitive relationship with phytoplankton for nutrients and light (Scheffer et al. 1993) or the production of allelopathic substances that can inhibit phytoplankton growth (Korner & Nicklisch 2002; Hilt & Gross 2008), especially of some cyanobacteria (Nakai et al.2001, 2012). However, these mechanisms are only effective below the “critical turbidity” threshold (Scheffer et al.1993), above which light limitation prohibits macrophytes growth and can lead to macrophyte die off, which marks the transition to a turbid water state (Schefferet al. 1993; van Nes et al. 2007; Kéfi et al. 2016; Yamamichi et al. 2018). In our experiment, macrophytes died out and did not re-establish after the final pulse of the first nutrient addition (October 10th 2016) until the following spring, which we confirmed by visual inspection of all ponds in March 2017. Therefore the observed differences between treatments with and withoutMyriophyllum can only be explained by the legacy of their prior impact throughout the summer and fall. Macrophytes also affected the dynamics of dissolved organic matter (Fig. 2): fDOM increased more rapidly and to higher levels in both M and MD treatments than in ponds without Myriophyllum (C and D). This was expected, asMyriophyllum is known to be a producer of a wide range of organic substances, including allelopathic chemicals (Catalán et al.2014; Reitsema et al. 2018).
The presence of Dreissena alone lead to the expected mediation of phytoplankton biomass, relative to the control without foundation species during parts of the first, and, by tendency, also throughout the second period of nutrient addition. Filter feeding organisms like Dreissena can remove large quantities of algae and suspended materials from the water column, which can help stabilizing aquatic ecosystems in a clear water state, even when the nutrient input is high (Gulati et al.2008; McLaughlan & Aldridge 2013). In this context, Dreissenahave higher persistence than Myriophyllum , because they are not limited by increasing turbidity like macrophytes. It has been shown that population growth of mussels can be very high in eutrophic lakes (Karatayev et al. 2014a; Strayer et al. 2019), if sufficient amounts of hard substrate are available (Ibelings et al. 2007; Fishman et al. 2010). In such cases, Dreissenacan not only affect water clarity and nutrient cycling, but also directly lead to shifts in the composition of the phytoplankton community towards a higher proportion, in some cases dominance, of cyanobacteria like Microcystis (Vanderploeg et al. 2001; Bierman et al. 2005; Fishman et al. 2010) . Dreissenacan selectively reject particles as pseudofeces that bypass the digestive tract, thus releasing less palatable particles like cyanobacteria back to the environment (Vanderploeg et al.2001). If this loosely consolidated substrate contains viable cyanobacteria, these cells can resuspended in the water column while other phytoplankton species are absorbed by the mussel.
The observed non-additive antagonistic effect of Myriophyllum and Dreissena coincided with a dramatic shift towards cyanobacteria that occurred when both macrophytes and mussels were present (Fig. 2). As found by Narwani et al. (2019), who determined phytoplankton community composition from pond water samples taken at regular intervals in the first year of the study, the small cyanobacterium Synechococcus was dominant when bothMyriophyllum and Dreissena were present in a pond. In a parallel laboratory experiment, Narwani et al. (2019) tested how the presence of allelochemicals (“Myriophyllum -tea”) orDreissena , alone and in combination, affected the relative concentration of two species of microalgae that were most dominant in the pond ecosystems (Lagerheimia sp. and Synechococcussp.). Similar to the dynamics observed in the pond experiment,Synechococcus increased in abundance relative to the green algaeLagerheimia when both Dreissena and allelochemicals were present. This suggests that a relative advantage of cyanobacteria in the presence of both foundation species, while other taxa in the community experienced stronger negative effects, may have contributed to the shift of phytoplankton communities toward cyanobacteria, resulting in an overall increase in phytoplankton biomass (Narwani et al. 2019).
Throughout the study, we found strong effects of the nutrient disturbances on the dynamics of ecosystem metabolism, which varied among treatment combinations of the foundation species. In both periods of disturbance (i.e. Figure 4, Phase 1 and 3), the dynamics of ecosystem metabolism such strong evidence of non-additivity, whereas in the intervening period (Figure 4, Phase 2) the differences among treatments were more subtle, and the overall patterns were driven by seasonality. For example, all metabolic rates increased over the spring until the middle of June, and then decreased until the final nutrient addition at the beginning of Phase 3 in October. Moreover, in the MD treatment the CV of GPP was often higher than the other treatments during the period when seasonality and weather events likely dominated the dynamics. CV is a commonly used metric for early warning sign for shifts in ecosystem state (just like AC, GEV, and skewness - see supplement), and the increase towards the end of Phase 2 hints that the ecosystems might respond differently to the impending pulse disturbance in Phase 3 (Figure 4). This suggests that high frequency time series might provide insight into how ecosystems will respond to disturbance. Following the final nutrient addition, all ecosystems containing foundation species (D, M and MD) showed significantly lower GPP and NEP, but higher R. This could be because chlorophyll concentration in the control ponds without foundation species continued to increase throughout the winter 2017/2018, whereas DOM concentration in all other ponds decreased (Fig. 2). As a consequence, higher productivity from phytoplankton in the control ponds and higher respiration from DOM breakdown in all other ponds may be responsible for the observed divergence in metabolic patterns towards the end of the experiment.
Multiple lines of evidence suggest that non-additive interactions between Myriophyllum and Dreissena strongly affected ecosystem dynamics in ponds experiencing progressive nutrient perturbations. This was especially visible in the phytoplankton communities: the presence of both Myriophyllum andDreissena led to a higher algae biomass relative to control, instead of a decrease when only one species was present in the ponds. This demonstrates how a non-additive, antagonistic interaction between two foundation species can have dramatic effects on the ecosystem, by providing an opportunity for a third species, in this case cyanobacteria, to dominate the community. Ecological synergies following ecosystem perturbation are a known, but not well researched phenomenon (Suttle et al. 2007; Darling & Côté 2008; Thompson et al. 2018). In some cases it may be difficult to uncover the effects that non-additive species interactions have on ecosystems, particularly when they are only expressed under disturbance conditions: in our experiment, the phytoplankton biomass decreased again after we ceased the nutrient additions. Nevertheless, the ecological mechanisms underlying these effects might persist over time, even though the dynamics are not evident during times of no disturbance (e.g. Phase 2). In our study, even after perturbing the ecosystems a year later with a single strong pulse of nutrients, the effect was stronger than during the first addition, indicating that the non-additive effects of species interactions can persist over time in a repeatable way.