Impact of herbivory and competition on lake ecosystem structure: underwater experimental manipulation

Two basic ecological relationships, herbivory and competition, distinctively influence terrestrial ecosystem characteristics, such as plant cover, species richness and species composition. We conducted a cage experiment under natural conditions in an aquatic ecosystem to test the impacts of two treatments combined in a factorial manner: (i) a pulse treatment – removal of dominant competitors among primary producers (macroalgae Chara sp. and Vaucheria sp.), and (ii) a press treatment – preventing herbivore (fish, crayfish) access to caged plots. The plots were sampled once before the treatments were established and four more times within two years. Both treatments had a significantly positive impact on macrophyte cover and species richness and changed the macrophyte species composition. The effect of the macroalgae removal was immediate with the highest species richness occurrence during the first post-treatment monitoring, but the positive effect vanished with time. In contrast, preventing herbivore access had a gradual but long-lasting effect and reached a more steady-state over time. Two of the most common species showed contrasting responses, the palatable Potamogeton pectinatus was most supported by caging, while the distasteful Myriophyllum spicatum preferred open plots. Our findings may be applicable during the revitalisation of aquatic ecosystems that aims to increase macrophyte biodiversity.


Results
Regardless of the treatment, all plots underwent pronounced successional dynamics that was the same as the rest of the vegetation in the lake. This was marked by the significant effect of time in all the statistical analyses. Nevertheless, the successional dynamics always changed due to the effect of the experimental treatments.
Macrophyte cover. Hereafter, macrophyte cover refers to the cover of aquatic plants other than the manipulated dominant species of primary producers (i.e., macroalgae Chara sp. and Vaucheria sp.). Both treatments, i.e., macroalgae removal and prevention of herbivore access, had a significant positive effect on the macrophyte cover that changed over time; this included the significant main effects of both treatments and that of their interaction over time ( Table 1). The increase in cover differed markedly in each treatment. Macroalgae removal resulted in an immediate significant increase in macrophyte cover during the first period (time 1; increase from 0% to 15%), followed by a gradual decrease to 1% over time ( Fig. 1 and Table 2). The removal was successful, and the cover of dominant macroalgae decreased, on average, from 100% to 3% after the treatment; however, the cover gradually increased to 78% in time 4 (Fig. 2). Preventing the access of herbivores resulted in a slow gradual increase, from 0% to 11%, in macrophyte cover in time 3. The cover subsequently decreased to an average of 3% in time 4 (Fig. 1). The interaction of the two treatments was not statistically significant (neither itself, nor in the interaction with time); thus, the impacts of the two treatments on the macrophyte cover was additive on the log scale (Table 1). This might be interpreted as the two treatments having independent effects. In plots subjected to both caging and removal, the macrophyte cover increased in time 1, as it did in plots with only macroalgae removal treatment. In addition, a subsequent gradual decrease in macrophyte cover was observed, similar to the decrease seen in the cage-only plots (Fig. 1).
In the control plots, the changes were the least evident, and the macrophyte cover was the lowest (Fig. 1, Table 2). The explained variability in ANOVA showed that, macroalgae removal initially had a larger effect than did the prevention of herbivore access (Fig. 3a); however, the effects of macroalgae removal quickly faded. The decrease in the effect was apparently caused by the return of macroalgae to the experimental plots (Fig. 2). The prevention of herbivore access reached its maximum effect in time 2, and then the effect slowly decreased but remained significant until the end of the experiment. In contrast, the highest effect of the macroalgae removal was measured in time 1 and then decreased quickly (Fig. 3a).

Species richness.
As with macrophyte cover, species richness was positively affected by both treatments (i.e., both had a statistically significant interaction with time), and their interaction was not significant, neither by itself, nor in its interaction with time (Table 1, Fig. 4). In addition, similar to macrophyte cover, the  effect of removal was very pronounced at the beginning of the experiment, but this effect disappeared in the last two observations; in contrast, the effect of caging was rather negligible at the beginning of the experiment but appeared as significant in the last observation (Fig. 3b). In the case of macroalgae removal, species richness increased from one species in time 0 to five species in time 1 (each species was present in at least one of six plots for each treatment), and species richness subsequently decreased to one species per plot in time 4. The prevention of herbivore access resulted in a gradual increase in species richness, followed by a more steady-state (Fig. 4). The interaction of the two effects was not significant ( Table 1). The lowest species richness was generally observed in the control plots (Fig. 4).

Species composition.
In total, seven macrophyte species and two macroalgae species were observed in the experimental plots. The species composition showed pronounced dynamics, i.e., both common successional dynamics and differential responses to the treatments. The first PRC (principal response curves) axis ( Fig. 5) was determined by the two most common species and their differential responses to caging; in contrast, P. pectinatus was strongly supported by the exclusion of herbivores, and Myriophyllum spicatum preferred the open plots. The PRC diagram (highly significant, pseudoF = 52.3, p = 0.0002, Fig. 5) clearly demonstrated the responses to experimental manipulations. While the response to macroalgae removal was immediate and pronounced, the effects vanished over time, as can be expected for a "pulse type" treatment; however, the effects of caging were gradual but rather long-lasting, corresponding to the "press type" treatment. This also corresponds to the amount of explained variability in individual RDA for the individual times ( Fig. 3c, Table 2). At the beginning of the experiment (i.e., time 1), the community composition was determined mainly by macroalgae removal (Fig. 5: the PRC curves in time 1 differ for removal, and in the RDA results for time 1, Supplementary Fig. S1, the first axis was determined by removal); additionally, most species responded positively to removal (with the exception of Potamogeton crispus). At time 2, caging was more important (the first RDA axis for times 2-4 were determined by caging). However, in time 2, both the effects were significant, and we can see a predominance of most of the species, i.e., M. spicatum, Myriophyllum verticillatum, Elodea canadensis and P. crispus predominated in the plots where macroalgae were removed. Interestingly, P. pectinatus responded positively to caging beginning in time 2, whereas the predominance of M. spicatum in the open plots started only in time 3. Only two species of macrophytes occurred in plots with no treatment: M. spicatum and P. pectinatus. Three species occurred in plots with prevention of herbivore access: M. spicatum, P. pectinatus and P. crispus. These three species, plus M. verticillatum, E. canadensis and Batrachium trichophyllum, occurred in plots where macroalgae were removed. In plots with both treatments, all mentioned species except E. canadensis were observed, and additionally, Potamogeton pusillus appeared.

Discussion
Our study demonstrated how competition and herbivory affected the structure of the aquatic ecosystem in terms of affecting the macrophyte cover, species richness and composition of macrophytes. The results supported our hypothesis that the pulse treatment (i.e., macroalgae removal) would result in a marked increase in the macrophyte cover and macrophyte species richness in the experimental plots; however, the initial effects quickly diminished and finally disappeared. The increase was apparent mainly during the first time period (i.e., time 1). Subsequently, the macrophyte cover decreased because macrophytes were gradually replaced by macroalgae. The species richness also started to decrease in time 1. A similar trend was observed during the succession stages of a drying lake in France, where the greatest species richness was observed during the first year 42 . A lake, like an island, is a semi-isolated ecosystem with a limited possibility of species dispersal from the outside 49 . Thus, a few competitively successful species may colonise the disturbed habitat after a short time, resulting in highest species richness occurrence during the initial succession stage 50 ; however, species richness subsequently decreases due to competitive exclusion. Macrophyte species that occurred only in plots after the pulse treatment -macroalgae removal (i.e., after the elimination of competition) included M. verticillatum, E. canadensis and B. trichophyllum. These three species were scarce, and each of them accounted for less than 1% of the total vegetative cover in the lake 17 . In addition, M. verticillatum is included on the IUNC red list and is classified as vulnerable (A2c). Although E. canadensis is an invasive species and is spreading worldwide 51 , the conditions in Milada Lake, such as trophy, water chemistry, sediment composition and hydro-morphological parameters [6][7][8] , are probably sub-optimal for its growth. The next species, B. trichophyllum, was observed in Milada Lake for the first time. Its occurrence demonstrated the apparent impact of disturbances on the community of primary producers in the lake. In terms of the condition of our study site, all the species mentioned above represent species that have the ability to quickly colonise and have low competitive ability 36 . Considering the isolation of Milada Lake, we may exclude hydrochory (i.e., water dispersal through ditch networks) in terms of colonisation by the new species B. trichophyllum. We assumed that waterfowl may have had an important impact on the dispersal of macrophytes to our locality, but we have no clear evidence to support our theory. Nevertheless, this hypothesis has been generally accepted [52][53][54] . Specifically, internal dispersal, which refers to the eating and excretion of seeds by waterfowl, has a more important role than does external dispersal, which refers to seeds that have attached to the body surface, a process that has been overestimated in the past 55 . However, we admit that the species could have already been in the lake, but if so, it was present only at very low densities that were not recorded during our regular monitoring. Our hypothesis that the decrease in herbivory caused by caging will positively affect both macrophyte cover and species richness was confirmed. The effect appeared much later and was particularly pronounced in terms of species composition, which increased with time. The results support the latest findings that herbivory in aquatic ecosystems is noticeable. Until the 1990s, herbivory on aquatic vascular plants was considered relatively unimportant; however, new findings show that 48% of plant biomass may be removed due to herbivory. This is, in fact, a pressure that is five to ten times greater than that reported in terrestrial ecosystems 56 . According to Wood et al. 57 , changes in plant abundances were reported at relatively low herbivore densities, suggesting that, in aquatic systems, greater herbivore densities overwhelm plant compensatory growth responses. The more intensive utilisation of plant biomass in aquatic ecosystems is probably due to low C: N ratio (i.e., the high content of nitrogen) in aquatic ecosystems 56 . We focused on the level of impact of herbivory on the macrophyte cover and species richness. Our complex monitoring of fish populations at the study site showed that the abundance of potential herbivorous fish able to affect the macrophyte community was 9-15 kg ha −1 (J. Peterka et al., unpubl. data). Rudd has the greatest impact in the studied lake 17,58 . Thus, rudd poses the most herbivorous pressure on the macrophytes in this experiment. Other potential herbivores include roach (Rutilus rutilus) 58 and spiny-cheek crayfish (Orconectes limosus; based on our observations), which are abundant in the lake but present a lower impact. We may exclude the impact of herbivorous waterfowl 52 , particularly the mute swan (Cygnus olor) and common coot  (Fulica atra), because the experiment occurred at the depth of 3-4 m 59 . In addition, a recent meta-analysis by Wood et al. 57 . referred to substantial between-taxa differences in the effects of herbivores on the abundance of freshwater and marine macrophytes. Fish have large impacts on macrophytes, while insects and birds have relatively low impacts on macrophytes. The reason for these differences may be the mobility and habitat preferences of each of these groups. Fully aquatic species, which permanently live underwater, have been shown to produce the greatest impacts on aquatic plants and are often considered to be ecosystem engineers 60 . Two species from our herbivores were invasive species in certain parts of the world, e.g., the spiny-cheek crayfish in Europe 61 and rudd in North America and New Zealand 18,20 . Therefore, it is advisable to have knowledge about their structuring role in an aquatic ecosystem.
The decrease of herbivory caused by caging also affected the species composition. This result is in accordance with the theory that herbivory has an important impact on the succession of macrophytes 56 . At our study site, the highest expansion was recorded for P. pectinatus, while the lowest cover was recorded for M. spicatum. These two species were the most common macrophytes in the experimental area as well as in the entire lake. The results clearly showed a preference for P. pectinatus by herbivores, but M. spicatum was ignored; furthermore, M. spicatum predominated in the open plots. The predominance of P. pectinatus in the caged plots appeared earlier than did the predominance of M. spicatum in the open plots. This suggested that M. spicatum, as a competitively weak species, was resistant to herbivory and took advantage of the open plots that suppressed the other species, particularly P. pectinatus. This result is in agreement with previous studies that showed that P. pectinatus was readily utilised by rudd 17,20,23 . The distaste for M. spicatum was probably caused by their high phenolic concentration, which is used as a chemical defence against herbivory 23 . The low number of M. spicatum in closed plots was probably due to its low competitiveness compared to P. pectinatus and other macroalgae that are more successful when herbivory is absent. This result demonstrates a typical trade-off, i.e., species with a chemical defence against herbivory are bad competitors and vice versa 1 . Like P. pectinatus, P. crispus was also positively affected by the "prevention of herbivore access" treatment but in considerably lower abundances. This result is in accordance with the claim that Potamogeton species generally represent a genus preferred by herbivores 17,20,32 , probably due to their low chemical defences 23 . P. crispus tries to defend itself with tough and partially spine-like leaves, but it seems that this type of defence is not very effective.
We found no significant interaction between the two treatments; thus, we can conclude that the effects of herbivory and competition were independent in our case (as no interaction means additivity on the log scale, for both species richness and macrophyte cover; additionally, the caging means the same percentage increase in both variables in the control and removal plots, and removal means the same percentage increase in the caged and open plots). Nevertheless, a positive effect of both treatments together was recorded for P. pusillus and B. trichophyllum. Both species were rare in the experiment as well as in the entire study lake. This indicated favoured utilisation by herbivores (results for similar species in 20,32 ) and highlighted their low competitiveness due to the suboptimal conditions in the lake, which was mainly represented by low nutrient availability 6,7,10 . However, due to the scarcity of these two species, they had very small effect on the results of the statistical tests.
We have clearly demonstrated that both herbivory and competition were strong ecological forces that shaped the macrophyte community composition. The results of our study may also be applied in revitalisation or conservation projects conducted in aquatic ecosystems. The regular disturbance of a lake bottom in a restricted area covered by macrophytes may lead to an increase in the biodiversity and preservation of rare species in the community, either by the colonisation of the uncovered bottom or by strengthening the diaspore production in the system 42,62 . A similar effect has been observed in various terrestrial ecosystems 63 . In contrast, human-induced disturbances of seagrasses in marine ecosystems may have a negative impact on sensitive species 64 . Thus, the controlled disturbances should be evaluated individually. Our results implied that disturbances such as those studied in our experiment will increase the biodiversity in an aquatic ecosystem. However, this approach cannot be applied without precautionary measures, as not all species would benefit from similar interventions. The treatment should be mainly performed in a limited area and should be continued after the pilot data have been evaluated. Though it is well-known that the species composition and quantity of the fish community affect the water quality 65 , we have newly demonstrated that vertebrate herbivory can decrease the quantity and species richness of aquatic macrophyte communities. We have also demonstrated the effect of macroalgae competition on macrophytes. Thus, management that decreases these effects might lead to the increase in biodiversity and may even promote some endangered macrophytes species.

Methods
Study site. The study was conducted in the newly created, opencast mine Milada Lake (50°39′N, 13°58′E; first treatment, the second treatment, both treatments, or no treatment was applied. Each combination was replicated six times, and the plots were completely randomised (with restrictions to prevent the clumping of identical treatments in space; Fig. 6c). We also attempted a third treatment, i.e., transplanting two bunches of M. spicatum and two bunches of P. crispus to the plots, but the plants did not establish (i.e., transplants were seen dead in time 1); thus, this third treatment was disregarded. The potential herbivores were rudd, roach and spiny-cheek crayfish. The abundance and biomass of rudd was 7 ind. ha −1 and 3 kg ha −1 , respectively, in 2014, and 91 ind. ha −1 and 7 kg ha −1 , respectively, in 2015. The abundance and biomass of roach was 149 ind. ha −1 and 12 kg ha −1 , respectively, in 2014, and 14 ind. ha −1 and 2 kg ha −1 , respectively, in 2015 (J. Peterka et al., unpubl. data). Both fish species were concentrated in the littoral section of the lake, which also has high macrophyte occurrence (based on our observations). The plots were initially sampled before the experimental treatment and then four times after the treatment, i.e., two samples in July and September in both 2014 and 2015, respectively. The cover of all aquatic plants and algae was visually estimated by two independent SCUBA divers in each plot. The plot was visually divided into four 1 × 1-m subunits, and the cover was estimated in given subunits and averaged by each diver; finally, the mean cover obtained from both divers was used as the final value. The estimation of the macrophyte cover varied between the divers by 1.8% (from 0 to 8%). The water transparency varied from 3 to 6 m during the experiment. To prevent potential shading, the microalgae and sediments were regularly removed by a small broom from the cages, particularly from the lid of the closed cages.
Statistical analysis. The macrophyte community was characterised by species richness, macrophyte cover (univariate response variables), and species composition. The univariate variables (i.e., species richness, macrophyte cover and macroalgae cover) were analysed using repeated measures ANOVA. The data were log(x + 1)-transformed before analysis to improve normality and homoscedasticity. Because the effect of both treatments changed over time, we conducted separate analyses (i.e., separate two-way ANOVAs) for each individual observation. We also presented the amount of explained variability by each of the main effects at each time point (the amount of explained variability was calculated as SS effect /SS total for each treatment). Because it was a manipulative experiment, the explained variability can be considered a measure of strength of the effect of manipulation, and the statistical significance is proof of causality. The univariate analyses were conducted using Statistica 12 (StatSoft Inc.).
The species composition was analysed in the framework of constrained ordinations. Because the data were in the form of repeated observations, we used PRC 67,68 . Because the PRC (highly significant) demonstrated that the effects of the two treatments changed considerably over time (similar to the univariate analyses), we used simple analyses of the effects of the two treatments (i.e., removal and caging) from individual observation times. In this case, variation partitioning 68 was used to quantify the independent effects of the two factors and to test for their significance. Because the two factors were orthogonal, the overlap of their two effects was, by definition, zero; thus, we were only interested in the size of the two effects. This procedure provided a multivariate equivalent of the ANOVA results.
The multivariate analyses were conducted using CANOCO5, and the significance of the effects was obtained using Monte Carlo permutation tests (in PRC, with a corresponding hierarchical permutation scheme), with 4999 permutations. Data availability. The dataset analysed during the current study is available from the corresponding author on reasonable request.