Herbivore exclusion promotes a more stochastic plant community assembly in a natural grassland

Both bottom-up (e.g. nutrients) and top-down (e.g. herbivory) forces structure plant communities, but it remains unclear how they affect the relative importance of stochastic and deterministic processes in plant community assembly. Moreover, different sized herbivores have been shown to have contrasting effects on community structure and function, but their effects on the processes governing community assembly (i.e. how they generate the impacts on structure) remain largely unknown. We evaluated the influence of bottom-up and topdown forces on the relative importance of deterministic and stochastic processes during plant community assembly. We used the data of a 7-year factorial experiment manipulating nutrient availability (ambient and increased) and the presence of vertebrate herbivores (> 1 kg) of different body size in a floodplain grassland in The Netherlands. We used a null model that describes a community composition expected by chance (i.e. stochastic assembly) and compared the plant community composition in the different treatments with this null model (the larger the difference, the more deterministically assembled). Our results showed that herbivore exclusion promoted a more stochastic plant community assembly, whereas increased nutrients played a relatively minor role in determining the relative importance of stochasticity in community assembly. Large herbivores facilitated intermediate-sized mammal herbivores, resulting in synergistic effects of enhanced grazing pressure and a more A cc ep te d A rt ic le This article is protected by copyright. All rights reserved. deterministic and convergent plant community assembly. We conclude that herbivores can act as strong deterministic forces during community assembly in natural systems. Our results also reveal that although largeand intermediate-sized mammal herbivores often have contrasting effects on many community and ecosystem properties, they can also synergistically homogenize plant communities.


Introduction
The assembly of species into a local community on ecological timescales is influenced by deterministic and stochastic processes and dispersal (Vellend 2010).
Deterministic processes are niche-related, include interspecific interactions and environmental forcing, and generally lead to convergent communities (little site-to-site variation in species composition) when environmental conditions are similar. Stochastic processes are neutral-related (Hubbell 2001), include colonization/extinction dynamics and ecological drift and can lead to divergent communities (high site-to-site variation in species composition) among otherwise similar localities (Chase et al. 2009, Chase andMyers 2011).
The consensus is that both types of processes act simultaneously while their relative importance usually depends on the environmental context (Leibold and McPeek 2006, Adler et al. 2007, Yang et al. 2013). The challenge is to develop a framework which explicitly includes stochastic and deterministic processes and explores which factors influence the relative importance of each (Chase 2007, Chase andMyers 2011).

Accepted Article
This article is protected by copyright. All rights reserved.
Analogously, there was also a long-lasting debate regarding whether top-down (i.e. herbivores and their predators) or bottom-up (i.e. nutrients, water, light) factors are the main structuring forces of plant communities, and it is now accepted that both forces simultaneously regulate the biomass and α-diversity of primary producers (Hillebrand et al. 2007, Gruner et al. 2008. Top-down and bottom-up factors in isolation have been shown to affect the similarity in species composition and the relative importance of stochastic and deterministic processes during community assembly (nutrients: Chalcraft et al. 2008, Chase 2010, predation: Chase et al. 2009, granivory: Germain et al. 2013, physical stress: Chase 2007, Guo et al. 2014. Indeed, community assembly is highly dependent on a multitude of bottom-up and top-down trophic influences (Caruso et al. 2011). However, how interactions between these processes regulate the outcome of determinism and stochasticity in community assembly is poorly understood and has not received much attention (but see Yang et al. 2013, Steiner 2014. Evidence from natural communities shows that divergence in species composition (which might imply higher stochasticity) may increase with increasing productivity or nutrient availability (Chase and Leibold 2002, Chalcraft et al. 2004, Chase 2010. However, which processes lead to these patterns remains largely unanswered (Steiner 2014). Available evidence suggests that divergence (and stochasticity) increases with higher nutrient availability due to priority effects (Chase 2010, Kardol et al. 2013, Steiner 2014. This occurs likely because higher productivity generally leads to a higher rate of population growth, which facilitates the preemption or modification of niches before the arrival of late-arriving species (Fukami 2015). If the most abundant species differ between plots just by chance, higher nutrient availability allows them to more easily outcompete subordinate species, magnifying the original differences in abundance (see Chase 2010).

Accepted Article
This article is protected by copyright. All rights reserved.
However, these bottom-up factors may be modified by top-down control. Particularly in terrestrial systems, it is well known that vertebrate herbivores influence local plant αdiversity in a variety of direct and indirect ways (Olff and Ritchie 1998), depending on ecosystem productivity or nutrient availability: at low-productivity sites (or unfertilized treatments), the impact of herbivory on α-diversity is negative, while it turns positive at high productivity sites (or increased nutrient treatments) (Bakker et al. 2006, Hillebrand et al. 2007), mainly by reducing competition for light (Borer et al. 2014). Furthermore, differentsized herbivores usually exhibit different feeding modes and preferences, that differentially affect α-diversity (i.e. intermediate-sized vertebrate herbivores usually show more unpredictable effects than large vertebrate herbivores; Bakker et al. 2006). In this context, Ritchie and Olff (1999) suggested that the combination of large-and intermediate-sized herbivores can have compensatory or additive effects on plant communities that likely depend on soil nutrients (additive herbivore effects on vegetation when soil nutrients are not limiting).
However, besides their effects on community composition and ecosystem functions, it is not clear if different-sized herbivores can also regulate the relative importance of nichebased vs neutral-based processes behind those effects along productivity or nutrient availability gradients. So far, experimental results showed that consumers (herbivores, predators) in isolation can increase the relative importance of deterministic over stochastic community assembly, because consumers reduce the number of species that could live in those habitats (Chase et al. 2009), particularly by removing species without traits to deal with consumption (Germain et al. 2013).
Considering all the above, the main goal of this study was to determine whether topdown and bottom-up forces interact to regulate the balance between stochastic and deterministic plant community assembly. More specifically, we investigated to what degree

Accepted Article
This article is protected by copyright. All rights reserved.
intermediate-and large-sized herbivores have a synergistic effect controlling the relative importance of stochastic and deterministic community assembly; and whether this effect of different sized herbivores depends on soil nutrients. We expect an interaction between nutrient availability and grazing, in which (1) nutrients will promote a more stochastic plant community assembly that can be overruled by grazing (i.e. nutrient addition in the exclosures will result in a more stochastic community, but with herbivory this effect will be lost), (2) the combination of intermediate-and large-sized herbivores will lead to a more deterministic plant community assembly, (3) particularly under increased nutrient availability due to increased grazing pressure.

Study site
In this study we used data from a long-term experiment (Bakker et al. 2006, Veen et al. 2010, Smit and Ruifrok 2011 that was conducted in a 50-ha floodplain within the JunnerKoeland area, a 100-ha nature reserve along the river OverijsselseVecht in the northeast of The Netherlands (52° 32' N, 6° 36' E). The aboveground annual net primary productivity is 489 ± 19 g m -2 (Bakker et al. 2006), which is intermediate in comparison with other natural grasslands (Borer et al. 2014). This area has been grazed by livestock for centuries, and during the course of this study the site was grazed by cattle (1 heifer per 2.5 ha from April to October; Bakker and Olff 2003). Other larger (> 1 kg) vertebrate herbivores that occur in the study site are European rabbit (Oryctolaguscuniculus), roe deer (Capreoluscapreolus), and brown hare (Lepus europaeus; Smit and Ruifrok 2011), the latter two in very low densities. Vegetation in the grazed lawns is dominated by Agrostiscapillarisand Festucarubra (Smit and Ruifrok 2011). In addition, many (> 40)

Accepted Article
This article is protected by copyright. All rights reserved. subordinate species also occur, with annual rates of colonization and extinction above 40 % (Bakker and Olff 2003).

Experimental design
To achieve our goals, we used data from a fertilization and herbivore exclusion experiment that was established in May 1994 in JunnerKoeland and that lasted until 2001 (seven years). Seven years is not enough to reach the climax of succession in temperate grasslands, where herbivore exclusion will likely lead to invasion by woody species and forests will establish (Olff et al. 1999). However, considering that our study is longer than others addressing related topics in similar systems (Germain et al. 2013, Yang et al. 2013, and the high rates of interannual colonization and extinction observed at our study site (Bakker and Olff 2003), we believe that the length of our study is consistent with our goals.
The experimental design consisted of five blocks with three 12 m x 12 m grazing treatments each: (i) unfenced areas accessible to all herbivores (hereafter "controls"); (ii) areas fenced to exclude larger herbivores (i.e. cattle), but allowing access to smaller species (i.e. European rabbits; hereafter "large herbivore exclosures"); and (iii) areas fenced to exclude all mammals > 1 kg (hereafter "full exclosures"), invertebrates and small rodents were not excluded. Large herbivore exclosures had two barbed wires at 0.5 and 1 m height.
Full exclosures consisted of 1 m high chicken mesh. Within each replicate of each of the three grazing units, two 2 m x 2 m permanent plots were established following a randomized design. One of these was fertilized and the other was left unfertilized (Appendix S1: Fig. S1).
Nutrients were administered twice per year from 1994 onward, once in mid-April and again in early May, at an amount of 15 g N, 5 g P and 10 g K m -2 year -1 . Nutrients were dissolved in 10 l water and evenly distributed over the plots. This nitrogen addition rate is above the atmospheric nitrogen deposition rates (Reay et al. 2008).

Accepted Article
This article is protected by copyright. All rights reserved.
The vegetation composition in all plots was recorded annually for the seven years (1995 -2001) during peak biomass (late July / early August) by visually estimating cover following the decimal Londo scale (Londo 1976). Aerial cover was estimated visually per species by the same person in each year. Here we focus on the last year, but information regarding previous years can be found in Appendix S2. Total cover could exceed 100% if species vertically overlapped. To evaluate whether fertilization changed the grazing pressure in control plots (i.e. accessible to all herbivores), in the summer of the fourth year (July 1998) we measured grazing signs in leaves by putting a measuring stick (50 cm) horizontally on the vegetation. We inspected one leaf every 5 cm along the stick, and recorded whether it was grazed or not. We repeated this procedure five times, and thus, we sampled 50 leaves per plot.
Additionally, in order to compare our system to others, we also looked at the separate and combined impacts of herbivory and nutrients on vegetation height, photosynthetically active radiation (PAR) interception and α-diversity. Our grassland showed responses to grazing and nutrients similar to many other grasslands worldwide (all the information regarding these and other analyses is available on the Appendix S2). Finally, we also measured rabbit dropping's rate to evaluate the potential facilitation from cattle to rabbits and substantiate the expected increased grazing pressure in control plots (Appendix S2).

Data analysis
We first explored the effects of grazing and nutrients on the variability in community composition, and then we explored if differences in variability between treatments were indicating differences in the relative importance of stochastic and deterministic processes in community assembly. Variability in community composition was evaluated using multivariate group dispersions (Anderson et al. 2006), based on the Bray-Curtis similarity

Accepted Article
This article is protected by copyright. All rights reserved. index applied to fourth-root transformed data (to reduce the weight of the most abundant species on the index of similarity) and with 9,999 permutations. We restricted plot permutations within grazing treatments, and grazing treatments within blocks to account for the spatial dependence of the sampling units. We performed this analysis using the betadisper and permutest functions of the of the vegan package (Oksanen et al. 2015) for R. We did the analysis separately for the first and last year of the experiment, to evaluate whether the differences observed at the end of the experiment were already present at the beginning.
To be able to disentangle whether differences in variability were the product of differences in the relative importance of stochastic and deterministic processes in community assembly, we followed a null model approach that takes into account species abundances.
Our approach is a slight modification (to consider species abundances) of the null model proposed by , which in turn is a slight modification of the Raup-Crick index (Raup and Crick 1979). We included relative abundance of species because many studies showed that this carries useful information for understanding ecological processes (Hillebrand et al. 2008, Winfree et al. 2015. To achieve this, we followed a probabilitybased randomization (see Mori et al. 2015 for a discussion about abundance-based null models), in which we defined four constraints: (1) the number of species in a given plot was the same as the actual number of species observed in that plot; (2) the probability of each species to be sampled was the proportion of plots where this species was actually observed; (3) the total abundance (i.e. cover) in each randomly generated assembly was the same as the actual total abundance in that plot; and (4) the abundance probability of a given species was the actually observed proportional abundance of that species across all plots.
For a given pair of plots (30 plots in total: 2 nutrient treatments x 3 grazing treatments x 5 blocks; a pair could belong to different treatments), each plant community composition was probabilistically generated 9,999 times. For each iteration, the Bray-Curtis dissimilarity

Accepted Article
This article is protected by copyright. All rights reserved. index between those two plots was calculated. Finally, the resulting metric was the proportion of iterations in which the calculated Bray-Curtis index was smaller or equal than the observed index (i.e. calculated from the experimental results). This index was then modified to range from -1 to 1, by substracting 0.5 and multiplying this difference by 2 , Stegen et al. 2013b. Hereafter we will refer to this metric as RC ab (for abundance-based Raup-Crick). The resulting metric indicates whether a pair of plots is less similar (approaching 1), as similar (approaching 0), or more similar (approaching -1), than expected by chance. As this index becomes less similar to 0 (either positive or negative), the relative importance of deterministic processes increases (because the community assembly is less likely explained by chance). In other words, the higher the deviance from 0 (purely stochastic assembly), the more deterministic is the community assembly. Values approaching -1 (more similar than expected by chance) reflect processes that make replicates to converge (e.g. low site-to-site variability in species composition due to environmental filtering), while values approaching 1 (less similar than expected by chance) reflect processes that make experimental plots to diverge (e.g. higher site-to-site variability in species composition than expected by chance, due to spatial heterogeneity). The size of the species pool plays an important role in the calculation of this index, and following others (Chase 2010, Stegen et al. 2013a we defined the species pool as the list of species found in the experimental plots throughout the experiment.
To evaluate the separate and interactive effects of herbivory and nutrients (fixed terms) on the relative importance of stochastic and deterministic processes, we developed mixed-effects models in which grazing units nested in blocks were considered as the random terms. We selected (from the dissimilarity matrix described in the preceding paragraph) the comparisons of each replicate of a treatment to all the other replicates of that treatment (n = 5 replicates per treatment, therefore each plot had 4 comparisons; e.g., 1 vs 2, 1 vs 3, 1 vs 4 and

Accepted Article
This article is protected by copyright. All rights reserved. 1 vs 5). Our dependent variable was the mean RC ab on a plot level, and thus we calculated the mean RC ab between those four comparisons. This technique was used by Koerner et al. (in press) to eliminate issues with autocorrelation (n=5 rather than n=10). We used the nlme (Pinheiro et al. 2016) package for R to perform the linear mixed effects analyses. Visual inspection of residual plots did not reveal any obvious deviations from homoscedasticity or normality. The final model was selected using the likelihood ratio test (always consistent with the AIC criterion), first evaluating the significance of higher order interactions and then the main effects if interactions were non-significant (Zuur et al. 2009). There were no cases in which candidate models had similar likelihood ratios. For multiple comparisons, we used the function glht from the multcomp package (Hothorn et al. 2008).

Results
The different treatments exhibited very similar spatial variability in species composition and abundance at the beginning of the experiment (F 5, 24 = 0.05, p = 0.998; Fig.   1). However, at the end of the experiment, spatial variability was higher in full exclosures, followed by large herbivore exclosures (not statistically different from any of the two other treatments), while controls were very homogeneous (F 5, 24 = 4.47, p < 0.001; Table 1; Fig. 1).
Furthermore, nutrient addition increased variability only within full exclosures. Despite the large differences in variability, there were relatively minor differences in the number of species found per treatment (i.e. all the species that appeared at least once among all replicates of a given treatment; ambient: control = 23, large herbivore exclosure = 19, full exclosure = 24; fertilized: control = 23, large herbivore exclosure = 18, full exclosure = 19).

Accepted Article
This article is protected by copyright. All rights reserved. factors did not interact (χ 2 = 1.50, df = 2, p = 0.47; Fig. 2). Communities grazed by all herbivores (control plots) were highly convergent and more deterministically assembled than the other treatments. Full and large herbivore exclosures were marginally different, with the former being slightly more divergent than the latter.
Moreover, although in full exclosures there was a consistent shift from relatively high deterministic to a highly stochastic assembly during the first four years (and stabilized since then; Appendix 2: Fig. S5), the temporal pattern of species dominance in the different plots showed no clear trajectories (i.e., there was no consistent change in community composition; Appendix 2: Fig. S6).

Discussion
In this study we experimentally evaluated the influence of bottom-up (i.e. nutrients) and top-down (i.e. herbivory) forces on the relative importance of deterministic and stochastic processes during plant community assembly. Our results revealed that herbivores and nutrients affected the relative importance of stochastic and deterministic plant community assembly. Specifically, we found that when both large-and intermediate-sized herbivores were absent, the resulting plant community was more stochastically assembled than when they were present, while nutrients played a relatively minor role. Below we discuss how these results relate to our initial hypotheses and the literature.

Accepted Article
This article is protected by copyright. All rights reserved.

Bottom-up forces governing the relative importance of stochasticity and determinism in plant community assembly
Whereas the impact of nutrients and herbivores on plant diversity is by now rather well-known, to our knowledge, there is little information regarding the joint role of nutrients and different sized herbivores regulating the processes behind plant community assembly.
We hypothesized that nutrient addition will promote a more stochastic community assembly which can be overruled by grazing, but this hypothesis was not fully supported by our results.
Instead we indeed found that herbivores overruled the nutrient effects (supported by the higher grazing pressure in control plots after fertilization), whereas in the full exclosures the plant community was more stochastically assembled than in the control plots, but the effect of fertilization was negligible. At least three alternative explanations emerge as to why nutrient addition did not lead to a more stochastic community assembly: (1) due to the intermediate productivity at our site, nutrient addition has no net effect on the divergence in species composition or on the balance between stochastic and deterministic community assembly (Chalcraft et al. 2008, Yang et al. 2013, Steiner 2014); (2) the lack of large differences in dominance between plots upon fertilization may account for the lack of an effect (Chase 2010, Kardol et al. 2013, Steiner 2014, and (3) the smaller effect of nutrients could also indicate that herbivores overruled the nutrient effects (supported by the higher grazing pressure in control plots after fertilization) and the effect of nutrients would only be visible in the full exclosures. However, this last possibility was only partially supported by our results: exclosures were, as expected, more stochastically assembled than control plots, but the effect of fertilization was negligible. The exclusion of large herbivores may be enough to induce a highly stochastic plant community assembly (supported in our case by RC ab values very close to 0), and thus, fertilization cannot cause a more stochastic plant community assembly. It could also be that large herbivores strongly reduce nitrogen

Accepted Article
This article is protected by copyright. All rights reserved. mineralization (~ 40% lower in control plots; Bakker et al. 2004), leading to some sort of fertilization inside the exclosures, and thus decreasing the impact of fertilization. Either one of these explanations or a combination of them could be taking place at our study site.

Top-down forces governing the relative importance of stochasticity and determinism in plant community assembly
We confirm our second hypothesis in which we predicted that the joint absence of large-and intermediate-sized herbivores will lead to a more stochastic plant community assembly. However, we reject our third hypothesis (stronger impacts of herbivores when nutrients are added) because the results only showed minor differences following nutrient addition. Again, several mechanisms can explain these results. (1) Consumers (including herbivores and predators) usually lead to a more deterministic community assembly (Chase et al. 2009, Germain et al. 2013 because only a subset of the species tolerates the stress of being consumed (reducing variability between local communities). However, this mechanism does not seem to explain our results, given that there were only minor differences in the total number of plant species found between treatments. (2) When several plant species can become dominant light competitors, the end product of this competition is likely stochastic local extinction rather than deterministic (Segre et al. 2014). For example, the loss of short plant species as a group is deterministic, but which of those species disappear or which tall plant species becomes dominant from the group of dominant light competitors can be stochastic. Our case fits with this mechanism given that we found the expected differences in plant species richness, vegetation height and PAR interception between grazing treatments at the plot level (Appendix S2). Furthermore, only two plant species reached cover > 20 % in a given control plot while seven plant species did it in full exclosures, and four of the five full exclosures did not share the same dominant plant species, while the same pair of plant species

Accepted Article
This article is protected by copyright. All rights reserved. always dominated in control plots (Table 1). (3) Alternatively, it might be argued that adding treatments always produces alterations in the community composition, that vegetation takes some time to adapt to the new scenario, and thus, that changes in plant community are reflecting altered conditions rather than the effect of our treatments. This is inherent of the method of herbivore exclusion, from which most of the knowledge of large herbivore impacts on the vegetation diversity, composition and dynamics originates (see for instance : Milchunas and Lauenroth 1993, Bakker et al. 2006, 2016, Borer et al. 2014. Given that in our study we are analyzing spatial variability (and not temporal variability), there is no reason to suspect that the different replicates should respond differently (unless there were underlying differences in environmental conditions, which was not the case here: Table 3 in Bakker et al. 2004). Hence we are confident in our results, and we conclude that the most likely mechanism behind a more stochastic community assembly when herbivory is relaxed is the second (stochastic local extinction).

Effects of different-sized herbivores
We specifically hypothesized that the simultaneous exclusion of large-and intermediate-sized herbivores will strongly favor a more stochastic community assembly, or alternatively, that the combination of all herbivores will result in deterministic plant community assembly, which was confirmed by our results. Large-and intermediate-mammal herbivores exhibit distinctive feeding modes and preferences (Semmartin andOesterheld 2001, Kuijper et al. 2008), and have contrasting effects on plant α-diversity Lauenroth 1993, Bakker et al. 2006) and plant colonization dynamics (Bakker and Olff 2003). Whereas these distinct roles of large-and intermediate-sized herbivores are acknowledged, it remains unclear whether they have compensatory or additive effects on grassland vegetation (Ritchie and Olff 1999). In our study, large herbivore grazing was the

Accepted Article
This article is protected by copyright. All rights reserved. most important driver of the more deterministic plant community assembly. At the end of the experiment, there were minor differences in the relative importance of stochasticity and determinism between both treatments without large herbivores (i.e. large herbivore and full exclosures). Nevertheless, intermediate-sized vertebrate herbivores played an important secondary role, because their absence (along with that of large herbivores) promoted a clear distinction between control plots (more convergent and deterministic plant community assembly) and full exclosures (more divergent and stochastically assembled). This was particularly due to a strong facilitative effect, where intermediate-sized mammal herbivores preferred to feed on plots grazed by large herbivores, imposing a stronger stress and thus promoting the more deterministic plant community assembly. Cattle reduced vegetation height, and short vegetation is preferred by rabbits, increasing overall grazing pressure on the control plots (Bakker et al. 2009). This facilitation from cattle to rabbits can also explain the intermediate pattern found in the large herbivore exclosures. After removing large grazers, while the vegetation is still short, rabbits can exert some control over the plants and retard the transition to the more stochastically assembled plant community. However, as vegetation grows and taller species become dominant, top-down control by rabbits decreases (as has also been observed in salt marshes with hares and geese: Kuijper et al. 2008) and eventually full exclosures and large-herbivore exclosures converge. Although in our system large-and intermediate-sized vertebrate herbivores were mainly represented by cattle and rabbits, we argue that this can be seen as a simplified model system of the diversity of vertebrate grazers and their functional roles. Most natural grasslands were once grazed by a diversity of large herbivores, but many of the native large herbivores went extinct in the past or are seriously threatened in the present (Dirzo et al. 2014, Ripple et al. 2015. The rabbits are representative of a guild of small-medium sized burrowing mammals, that includes prairie dogs, steppe pikas and ground squirrels, which

Accepted Article
This article is protected by copyright. All rights reserved. despite having important impacts on grassland ecosystems have not been frequently studied (see Davidson et al. 2010). Our study demonstrates the importance of interactions between large-and intermediate-sized herbivores, in this case resulting in altered grazing pressure and synergistic effects on grassland vegetation, emphasizing the importance of taking into account the role of herbivore diversity in studying herbivore impacts on grassland vegetation.

Conclusions
Overall, one of our three predictions was confirmed. In our study, nutrients did not play an important role regulating the relative importance of stochastic and deterministic plant community assembly (prediction 1), the combination of large-and intermediate-sized herbivores promoted a more deterministic plant community assembly (prediction 2), but this was not strongly modulated by nutrient availability (prediction 3). This study shows that, at least in intermediately productive grasslands, top-down control can be more important than bottom-up forces in regulating the relative importance of stochastic and deterministic processes shaping plant community assembly. The enhanced grazing pressure in nutrient enriched plots (capable of overruling bottom-up effects at the plot level) that we found has been documented many times before in herbaceous systems. In salt marshes, for example, several studies showed that herbivores entirely suppress the otherwise large effect of fertilization (Bertness et al. 2008, Alberti et al. 2010. Grazing lawns are another common example where top-down forces overrule bottom-up effects; herbivores aggregate in nutrient hot-spots, keep vegetation short and return more nutrients fueling this positive feedback (Augustine et al. 2003, Cromsigt andOlff2008). This study also reveals that the well-known effects of top-down forces overruling bottom-up effects on community and ecosystem properties might not regulate the relative importance of the processes behind community assembly. Finally, our results show that a higher diversity of consumers is more efficient in

Accepted Article
This article is protected by copyright. All rights reserved. overruling nutrient effects, and that this synergistic effect does not necessarily depend on nutrient availability. Hence, our study suggests that an impoverishment in the diversity of vertebrate herbivore communities may result in a lesser capacity of top-down control on plant community assembly, leading to less predictable local communities. Table 1: Percent cover per treatment and block for the 12 species that at the end of the experiment reached a cover >= 20 % (in bold). 1: Agrostiscapillaris; 2: Festucarubra; 3: Holcuslanatus; 4: Carexhirta; 5: Cirsiumarvense; 6: Rumexacetosella; 7: Rumexcrispus; 8: Urticadioica; 9: Sambucusnigra; 10:Stachyspalustris; 11: Arrhenatherumelatius; 12: Veronica chamaedrys. Grazing

Accepted Article
This article is protected by copyright. All rights reserved. given replicate, and the distance between any two points represents the difference between those two community assemblages according to the Bray-Curtis dissimilarity index based on fourth-root transformed data. Lines represent the confidence ellipse at the 0.95 level. The larger the ellipse of a given treatment, the greater variability of that treatment. . This metric ranges from -1 to 1, indicating whether a pair of plots are more dissimilar (approaching 1), as dissimilar (approaching 0), or less dissimilar (approaching -1), than expected by chance. Dashed horizontal gray line denotes a pure stochastic community assembly. Different letters denote significant differences between