You are on page 1of 10

tor Oikos 122: 967976, 2013

di
doi: 10.1111/j.1600-0706.2012.00114.x

s
E 2012 The Authors. Oikos 2012 Nordic Society Oikos
OIKOS
Subject Editor: Richard Michalet. Accepted 20 September 2012
Ch
oice

Bottomup and topdown forces structuring consumer communities


in an experimental grassland

Michael Rzanny, Annely Kuu and Winfried Voigt


M. Rzanny (michael.rzanny@uni-jena.de) and W. Voigt, Inst. of Ecology, Friedrich Schiller Univ., DE-07743 Jena, Germany. MR also at:
Chair of Land Improvement and Nature Conservation, Technische Univ. Dresden, DE-01737 Tharandt, Germany. A. Kuu, Tartu College,
Tallinn Univ. of Technology, EE-51008 Tartu, Estonia.

The interplay between bottom-up and top-down effects is certainly a general manifestation of any changes
in both species abundances and diversity. Summary variables, such as species numbers, diversity indices or
lumped species abundances provide too limited information about highly complex ecosystems. In contrast,
Synthesis

species by species analyses of ecological communities comprising hundreds of species are inevitably only snap-
shot-like and lack generality in explaining processes within communities.
Our synthesis, based on species matrices of functional groups of all trophic levels, simplifies community
complexity to a manageable degree while retaining full species-specific information. Taking into account plant
species richness, plant biomass, soil properties and relevant spatial scales, we decompose variance of abundance in
consumer functional groups to determine the direction and the magnitude of community controlling processes.

After decades of intensive research, the relative importance of topdown and bottomup control for structuring eco-
logical communities is still a particularly disputed issue among ecologists. In our study, we determine the relative role
of bottomup and topdown forces in structuring the composition of 13 arthropod functional groups (FG) comprising
dierent trophic consumer levels. Based on species-specic plant biomass and arthropod abundance data from 50 plots
of a grassland biodiversity experiment, we quantied the proportions of bottomup and topdown forces on consumer
FG composition while taking into account direct and indirect eects of plant diversity, functional diversity, community
biomass, soil properties and spatial arrangement of these plots. Variance partitioning using partial redundancy
analysis explained 2144% of total variation in arthropod functional group composition. Plant-mediated bottomup
forces accounted for the major part of the explainable variation within the composition of all FGs. Predator-mediated
topdown forces, however, were much weaker, yet inuenced the majority of consumer FGs. Plant functional group
composition, notably legume composition, had the most important impact on virtually all consumer FGs. Compared to
plant species richness and plant functional group richness, plant community biomass explained a much higher proportion
of variation in consumer community composition.

Natural communities are typically structured in concert interactions, there is still remarkably little consensus on the
by various biotic and abiotic forces (Hunter and Price relative roles of bottomup and topdown forces on multi-
1992). Biotic forces are further subdivided into topdown species community level (Boyer et al. 2003, Gruner 2004,
(predator controlled) and bottomup (resource controlled) Halpern et al. 2006). Additionally, abiotic factors modify
forces. Determining the relative importance of these forces interactions among species and so indirectly change the
challenged ecologists for decades (Hairston et al. 1960, strength of bottomup and topdown eects (Gripenberg
Hunter and Price 1992). A number of studies have focused and Roslin 2007, Alberti et al. 2010, Hoekman 2010).
on the relative roles of topdown and bottomup eects Biodiversityecosystem functioning research contributed
in dierent ecological systems and from dierent perspec- valuable insights into the question of how communities are
tives (Dyer and Letourneau 2003, Gruner 2004, Hoekman regulated and revealed the important role of plant diversity
2007, Gripenberg and Roslin 2007). Many experimental on arthropod diversity in grassland ecosystems (Koricheva
studies evaluated their relative impact on the level of indi- et al. 2000, Haddad et al. 2001, 2009, Scherber et al. 2010,
viduals (Kos et al. 2011), in very simple model ecosystems Cardinale et al. 2011). However, these studies consider uni-
(Hoekman 2010) or relied on experimental micro- or meso- variate descriptors such as species diversity and functional
cosms (Aquilino et al. 2005). However, due to the enormous richness and neglect dierences in community composition.
complexity arising from multiple and parallel direct inter- Community composition refers to the number, identity
action chains, accompanied by a complex array of indirect and abundance of species representing this community

967
analyzing the role of plant species richness, functional rich-
ness, and plant biomass for the composition of arthropod
functional groups and compare the impact of these variables
with the impact of the composition of functional groups
from all trophic levels. To this end, we used data from a
controlled biodiversity experiment (The Jena experiment)
minimizing environmental gradients inherent to previous
observational approaches.
Specically we asked: 1) which abiotic or biotic variable
groups best explain the composition of dierent functional
groups in arthropod communities? 2) how is the composition
of specic arthropod functional groups aected by the com-
position of the functional groups of other trophic levels, i.e. to
what extent do bottomup or topdown forces inuence the
composition of specic consumer functional groups? 3) is the
composition of dierent arthropod functional groups aected
by plant species richness, plant functional group richness and
Figure 1. Conceptual ecosystem diagram describing the hypo- community biomass and to what extent are these eects inde-
thesized and tested dependencies among the dierent ecosystem pendent from the eects of plant community composition?
components. Horizontal dependencies (within trophic level)
were considered for the carnivores to account for common trophic
relationships among the carnivorous functional groups. For func-
tional group abbreviations see Table 1. SR: Plant species richness,
Material and methods
FR: Number of plant functional groups.
Experimental design the Jena experiment

(Legendre et al. 2005). One of the very few studies using All data were collected in 2005 on 50 grassland plots
plant community composition to explain arthropod com- (each 400 m) of the Jena experiment (detailed experimen-
position in multispecies communities showed that plant tal design in Roscher et al. 2004). The experimental plots are
composition is consistently the most important predictor located in the oodplain of the river Saale at the northern
of community composition for several arthropod taxa edge of Jena, Germany (5055N, 1135E; 135 m a.s.l.).
(Schaers et al. 2008). Another study aimed to disentangle The plots were established from seeds on a former arable
the eects of productivity, plant diversity and plant composi- eld in spring 2002. To maintain target plant species
tion on arthropod abundance in mountain grasslands and composition, the plots were manually weeded (April and
likewise found plant species composition to be the most July) and mown twice a year (June and September), and
important predictor of arthropod abundance (Perner et al. remained unfertilized during the period of data collection.
2005). Productivity aected arthropod abundance to a The particular plant species mixtures were randomly selected
lesser extent, while plant diversity eects were not detected. from a species pool comprising 60 plant species typically
Moreover, there is evidence from a recent experiment in occurring in semi-natural mesophillic grasslands in the
saltmarshes showing strong eects of increased plant bio- surrounding region. We used 16 monocultures, 16 four-
mass production on arthropod community composition, species mixtures, 14 sixteen-species mixtures and four
independent of plant species richness (Wimp et al. 2010). 60-species mixtures for our analyses.
Ecological communities may be perceived as systems of All plant species were a priori assigned to functional
functional groups (FG), interacting with each other and groups: grasses (PGR, 16 species), legumes (PLG, 12 species),
with the abiotic environment (Fig. 1, Voigt et al. 2007, tall herbs (PTH, 20 species) and small herbs (PSH, 12 spe-
Blaum et al. 2011, Rzanny and Voigt 2012). Changes in the cies), based on the results of a cluster analysis comprising
composition of plant functional groups, plant species morphological, phenological and physiological traits (see
richness, functional richness, or community biomass are Roscher et al. 2004 for details). Species richness and the num-
expected to aect the composition of higher trophic levels, ber of functional groups per plot (functional richness) were
i.e. bottomup forces. In the same way, microclimate, varied as independently as possible (Roscher et al. 2004).
depending on soil conditions, and spatial relationships To take into account possible abiotic spatial patterns,
among the plots may aect arthropod composition. soil texture was determined using standard methods before
Herbivorous and detritivorous species may act as food setting up the experiment (Roscher et al. 2004) and soil
sources for carnivorous species and, hence, will aect their moisture (gravimetric water content) was measured in
composition. The carnivores in turn will aect herbivore and 015 cm depth (Oelmann et al. 2011).
detritivore composition, in fact, exerting topdown forces
on these trophic levels. Plant sampling
We consider these multiple and intricate relationships
using an ecosystem scale approach covering a multitude of Aboveground plant biomass was harvested between 30 May
factors potentially aecting arthropod composition at dier- and 8 June and again between 22 and 31 August (Weigelt
ent spatio-temporal scales (Fig. 1). We aim to resolve the et al. 2010). This was done by clipping the vegetation 3 cm
inconsistent relationships revealed by dierent studies above ground on four rectangles (each 0.20  0.50 cm)

968
positioned on predened locations. The coordinates of (HSU), seed predators (HSP), root feeders (HRF), special-
these positions were assigned randomly prior to the harvest ized chewers (HSC) and endophytics (HEP). The detriti-
in the inner area of each plot (at least three meters away vores are subdivided into endophytic saprophages (DES)
from the plot margins). Plant biomass was sorted to species and soil living saprophages (DSS). The carnivore trophic
level and dried at 70C for 48 h. The dry weight in g m2 level is formed by parasitoids (CPO), vegetation associated
of each species was calculated as the mean of the four suckers (CVS), ground living chewers (CGC), web spiders
subsamples per plot and was taken as a measure for plant (CWS), ground living suckers (CGS) and omnivores
community biomass per plot (Weigelt et al. 2010). (OMG) (Table 1). In combination with the plant func-
tional groups, which represent the producer trophic level,
Arthropod sampling we comprehend the whole community as a system of inter-
dependent functional groups (Fig. 1).
Arthropod abundance was estimated using two dierent
techniques. Two pitfall traps (4.5 cm diameter) were Spatial patterns
placed near the centre of each plot to estimate activity
density of epigeic, ground-associated species. The traps were Arthropod composition may be aected by spatial patterns
emptied every three weeks during the vegetation period across a variety of scales, ranging from broad geographical
(altogether six times). Only during the mowing periods, the gradients to ne scaled autocorrelation patterns. To account
traps were closed for about two weeks. Population density of for this, we used principal coordinates of neighbour matrices
vegetation-associated arthropods was measured using an (PCNM, Borcard and Legendre 2002, Borcard et al. 2004)
electric suction sampler in combination with a biocenometer to describe the spatial relationships among the plots. This
(bottomless, cube-shaped covering of gauze) covering an method allows extracting spatial patterns across the whole
area of 0.75  0.75 m. This method allows volume-specic range of all perceivable scales of a given spatial structure. The
sampling and yields highly comparable arthropod abun- resulting orthogonal eigenvectors could be used as additive,
dance data. Samples were taken on ve occasions from May scale specic spatial descriptors in a canonical analysis.
to October. The following taxa were sampled: Diptera, PCNM analysis is based on the computation of princi-
Hymenoptera, Coleoptera, Araneae, Opiliones, Heteroptera, pal coordinates of a truncated Euclidean distance matrix
Auchenorrhyncha and Orthoptera, representing a major describing the spatial relationships among the plots. The
component of the entire aboveground fauna in these geographical positions of each plot centre were used to calcu-
grasslands. With the exception of some predominantly late this distance matrix. The resulting PCNM axes were
parasitoid Hymenoptera, which were in most cases deter- named following their order (PCNM 1-PCNM 30) and
mined to the level of family or subfamily, all arthropods used as independent spatial variables.
were identied to species level. Only taxa with at least ten
sampled individuals were included in the analysis. Data analysis
All arthropod species were assigned to functional
groups (Table 1) based on cluster analysis of feeding and life All data were divided into ve variable groups (Fig. 2).
history traits such as body mass, trophic level, feeding Each functional group is represented as a species-by-plot
strategy, degree of specialization or aerial mobility (Rzanny matrix containing the biomass (plants) or abundance
and Voigt 2012). These functional groups reect the (arthropods) of each species (Fig. 2). The biomass-matrices
basic trophic community organization in herbivorous, detri- of the plant functional groups together with the experi-
tivorous and carnivorous functional groups (Fig. 1). The mentally manipulated variables plant species richness (SR,
herbivorous functional groups comprise plant suckers number of established plant species) and plant functional

Table 1. Functional group codes including species richness and total abundance of arthropods considered for the analysis.

Trophic level Code Functional group Important taxa No. of taxa Total abundance
Plants PSH small herbs misc. 12
PGR grasses Poaceae 16
PTH tall herbs misc. 19
PLG legumes Fabaceae 12
Herbivores HSU plant suckers Auchenorrhyncha, Heteroptera 44 9550
HSP seed predators Carabidae 25 4072
HRF root feeders Coleoptera 16 15 690
HSC specialized chewers Coleoptera 20 5076
HEP endophytics Diptera 26 2930
Carnivores CPO parasitic species Hymenoptera, Diptera 35 8196
CVS vegetation associated suckers Diptera 11 1580
CGC ground living chewers Carabidae, Staphylinidae 25 2313
CWS web spiders Araneae 33 4860
CGS ground living suckers Araneae 29 11 225
OMG omnivores misc. 28 12 514
Detritivores DES endophytic saprophages Diptera 5 382
DSS soil living saprophages Diptera, Coleoptera 25 3270

969
Figure 2. Overview of the analysis steps and derivation of the ve variable groups. Species-by-plot matrices of the plant functional
groups contain species-specic plant biomass, the species by plot matrices contain species abundance data. The variable groups plants,
carnivores, spatial patterns and soil were used as predictor variables for all consumer functional groups, while herbivore functional group
composition was a used as predictor variable for only the carnivore functional groups.

richness (FR, number of established plant functional groups) not suitable to detect such eects (Fig. 1). As a rst step, all
and complemented by total aboveground plant biomass variable groups had to be processed in dierent ways to
(biomass) formed the predictor variable group plants. The make them suitable for variance partitioning (Fig. 2). The
herbivore and detritivore functional group matrices together dierent data sets were log-transformed (all functional
formed the predictor variable group herbivores. They were groups matrices) or standardized by adjustment to stan-
combined in this group because both are placed as an inter- dard deviation (experimental variables, soil variables). The
mediate level between producers and carnivores (Fig. 1). functional group matrices were subsequently Hellinger-
Both act as food source for the carnivores and at the same transformed (Legendre and Gallagher 2001) to render
time consume dead or living plant biomass. Carnivore func- the species data suitable for redundancy analysis. All func-
tional group abundance matrices were considered as carni- tional group matrices were condensed to linear combina-
vores. All PCNMs with positive eigenvalues entered the tions using principal component analysis (PCA). However,
analysis as spatial patterns. The soil variables form the last we considered dierent numbers of ordination axes in
variable group soil. All variable groups entered the analysis the plant and consumer functional groups. Each plant
separately and all groups were independently subjected to functional group was reduced to six PCA-axes to ensure
variable selection in order to determine the overlapping parts that they comprise all potentially important information.
of variation accounted for by these dierent groups. When all variables forming these groups were analyzed
The consumer communities are allowed to colonize the simultaneously in a global redundancy analysis, extracting
plots freely from the surrounding area and are subject more than six axes substantially increased multicollinear-
to environmental lters and particular dispersal abilities of ity among the predictor plant functional groups (e.g.
the component species, which may result in positive or neg- highest variance ination factor [VIF] when 6 axes per
ative spatial autocorrelation patterns. In contrast, the plant functional group were used: VIF  2.9). The same proce-
composition of each plot is determined by the experimental dure was applied to consumer functional groups. However,
design and any species, which is not part of this planned the rst PCA-axes of the consumer functional groups
composition, is weeded out from the respective plot. There- showed particularly strong correlations (r  0.7) resulting in
fore, we explicitly did not consider any eects increased variance ination (for three axes per functional
on the plant communities, as our experimental design is group maximum VIF  8.9). Data exploration showed that

970
the rst axes were always strongly correlated to the commu- Table 2. Structure of the different pRDA-models explaining the inde-
nity biomass of the plots. Multicollinear predictor variables pendent fraction of variation in the composition of (a) herbivorous/
detritivorous functional groups (FG) and (b) carnivorous functional
within a variable group would reduce the interpretability groups. All respective variables were only included in the pRDA
of the variable selection results. The selection of a correlated if they had a signicant effect on the particular functional group
variable may obscure the causal eect variable within a composition. Abbreviations: SR: plant species richness; FR: plant
group of variables (Graham 2003) and therefore, impede functional richness.
the potential to reveal mechanistic explanations. Therefore, Predictor variables Control; effects partialled out in pRDA
we statistically controlled for the eect of community (a) Herbivorous/detritivorous functional groups
biomass on the PCA scores of all consumer functional Plant FG carnivore composition, soil structure, spatial
groups through the calculation of partial principal compo- composition patterns
nent analysis (pPCA), i.e. community biomass was Biomass, SR, FR plant composition, carnivore composition,
included as a covariable when calculating the PCA scores. soil structure, spatial patterns
Carnivore FG plant composition, soil structure, spatial
In doing so, we eectively partial out the eect of biomass composition patterns
on the composition of the consumer functional groups, which Spatial patterns plant composition, carnivore composition,
we intend to use as explanatory variables. On the other hand, soil structure
the eect of community biomass on functional group com- Soil structure plant composition, carnivore composition,
position enters the analysis separately via the variable group spatial patterns
plants. These adjusted pPCA-axes generally show lower cor- (b) Carnivorous functional groups
relations among each other (maximum VIF  5.4) and nota- Plant FG carnivore composition, soil, spatial patterns
composition
bly reduces correlations among the rst axes. However,
Biomass, SR, FR plant composition, herbivore composition,
including more than three pPCA axes per functional group carnivore composition, soil structure,
would again increase multicollinearity to critical levels. spatial patterns
To reduce the possibility that important information is Carnivore FG plant composition, carnivore composition,
missed by only considering three (consumers) or six (plants) composition herbivore composition, soil structure,
spatial patterns
composition axes, all species within one trophic level Herbivore FG plant composition, carnivore composition,
(i.e. without distinction in functional groups) were used to composition soil structure, spatial patterns
calculate as many pPCA (or PCA in the case of plants) axes Spatial patterns plant composition, carnivore composition,
as necessary to explain at least 80% of the total variation soil structure
within this trophic level. This set of axes was always analyzed Soil structure plant composition, carnivore composition,
spatial patterns
in the same way as the FG-resolved dataset to ensure that
no important information was lost through the inadequate
representation of variation in the ordination axes.
In the second step, all variable groups were subjected to a group compositions and the selected abiotic variable groups
forward selection procedure (Blanchet et al. 2008) except were determined separately through incorporating all
the experimentally manipulated variables SR and FR as signicant variables in turn as covariables in the pRDA
well as biomass, because we wanted to determine and (Table 2). Independent and shared eects of plant com-
compare their direct impact, relative to the other variables for position species richness, plant functional richness and
each functional group. This forward selection procedure community biomass were separately determined for each
involved two stopping criteria, which has been shown to consumer functional group. The composition of carnivore
greatly reduce type I error and reports reliable portions of functional groups was used as a potential predictor for
explained variance, two severe problems inherent to the classi- herbivore functional group composition and all con-
cal approach (Blanchet et al. 2008). Prior to the forward selec- sumer functional groups (excluding the respective response
tion, a global test (redundancy analysis) of the entire variable functional group from the predictor variable set) were used
group on the response functional group was performed. Only as predictors for the carnivore functional groups (Table 2).
when this test was signicant, we proceeded with forward The variable groups plants, spatial patterns and soil
selection. Selection of variables was stopped when either the were exclusively used in an explanatory role, while the com-
contribution of the included variable was not signicant or the munity composition of the arthropod functional groups were
explained variation, through the selected subset, exceeded the used both as predictor and explanatory variables (Table 2).
variation accounted for by the entire subset. The amount of explained variance depends, to a certain
As a nal step, the subsets of the selected variables were degree, on the number of species within this group. In
used for the partitioning of variation using redundancy order to compare the relative importance of the dierent
analysis (Fig. 2, Legendre et al. 2005, Peres-Neto et al. fractions, we always show and compare these fractions rela-
2006). We then calculated the unique and shared eects of tive to the total amount of variation explained by all
the dierent subsets on each functional group response variables together. We report unbiased adjusted R values
matrix and tested the unique fraction for statistical signi- ; Peres-Neto et al. 2006), which takes the number of
(R adj
cance using a permutation test with 9999 permutations included variables and the sample size into account and
(function anova.cca in package vegan, Oksanen et al. 2010). prevents an overestimation of the fraction of explained
Here, we tested the unique eect of the particular variable may take negative values if a variable explains
variance. Radj
group, independent of the eect of any other variables less variation than a random variable would explain. In
considered in the variation partitioning. Finally, the inde- this case, the fraction of explained variance was set to zero
pendent and shared eects of all selected functional (Peres-Neto et al. 2006).

971
Our approach is essentially correlative. The identied vari- individuals were the functional group with the lowest total
ance partitions, together with their associated p-values, repre- abundance (Table 1). The pPCA axes used as explanatory vari-
sent estimates as observational evidence that a certain source ables accounted for between 72% and 88% (six axes; plants)
of variation is important in structuring the composition of a and between 39% and 84% (three axes; consumers) of the
certain arthropod functional group. The variance partitioning original variation of each functional group. Using variance par-
approach used here does not explicitly address trophic cascades titioning, between 20.7% (CWS) and 43.6% (DES) of the
between the functional groups of adjacent trophic levels. total variation in FG composition could be explained (Fig. 3).
Instead, it quanties the strength and direction of dependency
between two functional groups. Species are simultaneously Effects of the variable groups on the composition of
subjected to topdown and bottomup forces (Hunter and functional groups
Price 1992). This idea is taken into account by using species
composition matrices as explanatory variables (Legendre and The variable group plants was generally the most impor-
Gallagher 2001, Halpern et al. 2006) and considering them tant variable group for the species composition of all con-
reciprocally as either dependent or independent variables. In sumer functional groups (Fig. 3). Plants explain a highly
contrast to lumped species abundance commonly used in uni- signicant fraction of variance in each functional group.
variate approaches (Koricheva et al. 2000, Haddad et al. 2001, However, strong dierences in variance attributed to these
Scherber et al. 2010), the composition of certain functional relative fractions existed between the dierent functional
groups at dierent trophic levels represents a much more groups. The average relative proportion of variation in her-
detailed measure of heterogeneity between the plots. We gen- bivore functional group composition due to plants was
erally did not account for multiple testing because of the 0.58  0.059 (arithmetic mean  SE), for the detritivore
exploratory character of this study and following the mathe- functional groups 0.62  0.109 and 0.49  0.042 for the
matical and logical arguments of Moran (2003). carnivore functional groups.
All statistical analyses have been performed using the The composition of herbivore/detritivore functional
statistical software R 2.12.0 (R Development Core Team) groups had signicant, unique eects on parasitoids and
using the packages vegan (Oksanen et al. 2010), and packfor vegetation associated suckers (Fig. 3) and some rather
(Dray et al. 2009). small and insignicant eects on web spiders, omnivores and
ground associated chewers.
The variable group carnivores explained signicant pro-
Results portions of variation in the composition of most herbivore
and detritivore functional groups and to a lesser extend of
In total, 81 658 individuals belonging to 322 arthropod taxa some carnivore functional groups (Fig. 3). The average rela-
were considered for the analysis. Root feeders, with 15 690 tive proportion of variation explained by carnivore functional
sampled individuals, formed the most abundant functional groups in the composition of the herbivore and detritivore
group, while detritivorous endophytics with 382 sampled functional groups was 0.16  0.046 and 0.07  0.031 in the

Figure 3. Relative fraction of variance explained (R adj) by the dierent variable groups for all functional groups. The small black bars at
the bottom represent the absolute total explainable variance for each functional group (axis title and scale at the bottom right).

972
carnivore functional groups. Again, there are strong dier- Effects of plant species richness, functional richness
ences among the dierent functional groups. and community biomass
The PCNM scores, signicantly aected the composition of
ground associated suckers, web spiders and root feeders, while Community biomass, plant species richness and plant func-
these eects were shared with other sources of variation in seed tional richness together explained signicant proportions
predators and omnivores. PCNM 7 and PCNM 1 were selected of variance in the composition of all arthropod functional
as signicant predictors in three functional groups (OMG, groups (except FR in DES; Fig. 4). Community biomass
HRF, CGS and HSP, HRR, CGS), followed by PCNM 2, accounted for a higher relative fraction of variance in commu-
which was selected two times (CWS, CGS). PCNM1 and nity composition of all consumer functional groups when
PCNM2 relate to broad-scale spatial gradients. PCNM7 repre- compared to plant species richness and functional richness.
sents a circular spatial pattern with similar plots within the outer The relative variation, attributed to community biomass, shares
margins and the interior plots of the experimental area, but a large fraction of explained variation with other variables and
great dierences between the outer and interior plots of the has a signicant, independent eect on the composition of two
experimental eld site (Supplementary material Appendix 1). carnivorous functional groups: web spiders and vegetation
Soil explained independently signicant fractions of associated suckers. Considering the sum of the independent
variation in the detritivorous functional groups (DSS: and the shared fraction, results in high proportions of variance
0.19; DES: 0.10) and also aected some functional groups explained in detritivorous (0.34  0.003) and carnivorous
of the other trophic levels (Fig. 3). functional groups (0.33  0.073) and only about the half in
herbivorous functional groups (0.16  0.011).
Compositional effects among functional groups Plant species richness showed a signicant, independent
eect on only root feeders. Functional richness on the other
Legumes represent the most important functional group, hand, showed a signicant, independent eect on the vegeta-
explaining the largest, highly signicant independent frac- tion associated suckers (Fig. 4). Nonetheless, both variables
tion of total explainable variance in composition through- share a large part of variation with other predictors. Jointly with
out all consumer functional groups (Fig. 4, Supplementary the shared fraction, species richness explained an average pro-
material Appendix 2). Here, the relative fractions of explained portion of 0.12  0.021 in the herbivore functional group
variance range from 0.16 (HSU) to 0.45 (OMG). Legume composition, 0.18  0.027 in detritivore functional group
composition explains an average fraction of 0.22  0.030 of composition and 0.16  0.025 in carnivore functional group
the total explainable herbivore functional group variation composition. Plant functional richness explains 0.11  0.017 in
and a markedly higher relative fraction in the detritivore herbivore functional groups, 0.10  0.065 in detritivore func-
functional groups (0.37  0.006) and the carnivore func- tional groups and 0.18  0.068 carnivore functional groups
tional groups (0.34  0.037). The composition of tall herbs when the shared fractions of variance are taken into account.
predict high and signicant fractions of variance in all her-
bivorous functional groups (average: 0.16  0.012) but less
so in only some carnivorous functional groups (0.027  0.014) Discussion
and in one of the two detritivorous functional groups (0.041)
(Fig. 4). Grass composition explained signicant indepen- Our results provide the rst general overview of the
dent fractions of variance for most functional groups, with relative importance of dierent biotic and abiotic forces in
the largest eect on plant suckers. A similar pattern appears determining multitrophic functional group composition in
for the small herbs, which show highly signicant impacts experimental grassland communities. We could experimen-
on some functional groups, but none on others (Fig. 4). tally support the view that plants (i.e. bottomup forces)
The herbivore functional groups explain much less predominantly determine consumer community composi-
variance of carnivore composition compared to the plant tion of various trophic levels (White 1978, Halaj and Wise
functional groups. They show strong uctuations regarding 2001, Kagata and Ohgushi 2006, Randlkofer et al. 2010).
their predictive power for the carnivores (Fig. 4). While
the endophytics explain signicant variations in the compo- Contributions of different variable groups to the
sition of parasitoids and vegetation associated suckers, composition of consumer functional groups
the specialized chewers were not selected in any of the for-
ward selection procedures. The compositions of plant suck- Our approach allows the disentanglement of the indepen-
ers and soil living saprophages account for a signicant dent and the shared eects of dierent variable groups. We
proportion of parasitoid composition. considered and quantied abiotic and spatial heterogeneity
The carnivore functional groups, again, show very dier- and showed that they play a minor role in structuring most
ent and specic contributions to the explained variance in consumer functional groups on the scale of our designed
the response functional groups (Fig. 4). However, there are experiment. The results show that although bottomup
only two cases where the composition of carnivore functional eects account for a major proportion of variation in the
groups explains a signicant fraction of variation in another composition of consumers from all trophic levels, likewise,
carnivore functional group. In contrast, all carnivore func- topdown forces exerted by carnivores have a signicant
tional groups explain a signicant proportion of variation in impact on most herbivorous functional groups (Hunter and
the composition of all but one herbivore functional group. Price 1992, Gruner 2004, Hoekman 2007).
Soil living saprophages is the only detritivorous functional Some recent studies found no support for the presence
group which is aected by a carnivore functional group (CPO). of topdown forces on consumer groups in experimental

973
Figure 4. Relative fraction of variance (R adj) in consumer functional group composition explained by the dierent predictor variables
(plant functional group composition, plant biomass (biomass), plant species richness (SR), number of plant functional groups (FR ),
herbivore functional group composition (only for carnivore functional groups) and carnivore functional groups). Dark colored fractions
of the bars represent independent eects, light colored fractions represent shared eects. Consumer functional groups, independently
and signicantly (p  0.05) aected by the respective variable are marked by an asterisk.

grasslands (Scherber et al. 2010), whereas, earlier experi- framework. This oers a higher resolution of the relation-
mental and observational studies corroborated the presence ships among functionally dierent species groups. Indeed,
of topdown forces structuring herbivore communities in there are marked dierences in the response of functional
grasslands (Koricheva et al. 2000, Gruner 2004, Perner groups (Fig. 2, 3). Some functional groups showed no
et al. 2005, Borer et al. 2006). Using path analysis, Koricheva signicant response to the composition of carnivore func-
et al. (2000) found that herbivore abundance was more tional groups (HRF, DES, CVS, OMG) indicating limited
strongly aected by parasitoid abundance than by plant topdown forces exerted on them. In contrast, carnivore
diversity, indicating topdown forces acting on the herbi- composition explained about 31% of the total explained
vores. A eld study, conducted in montane European variation of seed predators and about 25% of the total
grasslands, found that the highest proportion of variation explained variation of plant suckers (Fig. 3) and hence imply
in herbivores was explained by the (summed) abundances the presence of topdown forces. As indirect plant biomass
of carnivorous functional groups (Perner et al. 2005), like- eects on carnivores are accounted for by pPCA-axes and
wise indicating the presence of topdown forces. Unlike other potential common sources of variation are fully con-
the study of Koricheva et al. (2000) and Scherber et al. sidered using the variance partitioning approach, these
(2010), we used the exact species composition of trait-based eects are direct and not mediated by correlations with
functional groups as response variable in a multivariate other variables.

974
In contrast to a previous study by Crist et al. (2006), that arthropods from all trophic levels are much more inu-
spatial variation played a comparably minor role in structur- enced by species composition than by species diversity. This is
ing the consumer communities (Fig. 3). These dierences in line with other studies (Perner et al. 2005, Beals 2006,
were likely caused by methodological dierences and the Schaers et al. 2008, Vehvilinen et al. 2008, Zhang and
larger plot size of 20  20 m in our experiment. In our Adams 2011) and emphasizes that interactions between plants
analysis, the global test of all PCNMs excludes the spatial and consumers are determined primarily by the identity and
predictor variable set in many cases. The selected eigen- abundance of particular species (Randlkofer et al. 2010).
vectors PCNM 1 and PCNM2 represent large-scale spatial Plant biomass has stronger eects for most functional
gradients (Supplementary material Appendix 2). PCNM7 groups rather than plant species richness and plant functional
may reect an invasion pattern, showing the colonization richness (Fig. 4). Interestingly, Scherber et al. (2010) found no
process from the surrounding meadows is, at least for three signicant eect of plant biomass on any considered consumer
functional groups (CGS, OMG, HRF), still in progress. group, while plant species richness directly aected plant bio-
However, the nearly complete absence of signicant ne- mass and herbivore abundance. These authors considered
scale patterns indicates that the large plot size of 400 m summed abundances of three trophically distinct groups,
ensures the development of separate communities on each while we used thirteen species-resolved functional group
plot with autocorrelation playing a minor role in structuring abundance matrices as response variables. This raises the ques-
the consumer communities (Bommarco and Banks 2003). tion of the appropriate scale to tackle such questions and indi-
The impact of soil variables is most pronounced for cates that strong simplications may obscure eects that are
both detritivorous functional groups. This was expected and present on a ner scale. However, using a very similar approach
is in line with Perner et al. (2005) who identied soil and to the study of Scherber et al. (2010), Koricheva et al. (2000)
management variables together with plant composition as found plant diversity eects on arthropod abundance being
the most important predictor for the composition of detri- mostly indirectly mediated by changes in plant biomass.
tivorous functional groups.
Conclusions
The importance of legumes
Plant composition is by far the most important predictor
The importance of legumes in structuring consumer species for the composition of all consumer functional groups irre-
composition and interactions has been stressed earlier spective of their trophic level. Plant species composition
(Koricheva et al. 2000, Spehn et al. 2002, Scherber et al. integrates several abiotic and biotic environmental dier-
2006, Chen et al. 2010). Koricheva et al. (2000) found ences among ecological communities (Schaers et al. 2008)
the presence of legumes to be the major determinant of and hence is a very sensitive ecosystem descriptor. We
abundance for various arthropod groups in experimental strongly encourage ecologists to consider community com-
grasslands and stated that the eects of plant diversity on position when attempting to interpret the relative role of the
consumer species were mostly indirect and mediated dierent forces structuring ecological communities. Our
via plant biomass. The presence of legumes and plant results imply that invertebrate communities are mainly
diversity are inevitably linked to plant community biomass structured by bottomup forces, while additionally, top
(Spehn et al. 2002, Cardinale et al. 2011), but the relation- down forces aect nearly all herbivorous functional groups.
ship between these factors and the mechanisms causing com- We quantied and directly compared the impact of plant spe-
munity response to changes in plant composition are dicult cies richness, functional richness and community biomass on
to disentangle. In our analysis, this relationship is reected in the composition of functional rather than taxonomic groups.
a strong correlation between the rst ordination axis, repre- Our results reveal that community biomass is consistently
senting the strongest gradient in legume composition and more decisive for the species composition of all consumer
plant community biomass (Pearsons correlation r  0.78). functional groups than it is plant species richness per se.
This strong relationship mirrors the interdependence of these
two variables and renders impossible a statistical separation
of their relative contributions to changes in consumer func- Acknowledgements We thank all colleagues and people who
conceived and maintained the Jena Experiment and provided the
tional group composition. However, it has been shown basic plant and soil texture data, in particular C. Roscher, A. Weigelt
experimentally that increased primary production alone may and J. Baade. We are grateful to R. Mller who has a share in
change the structure and composition of terrestrial arthro- the eld data acquisition and kindly contributed data for some
pod communities in monocultures (Wimp et al. 2010). arthropod taxa, and Y. Oelmann and W. Wilcke who provided
data on soil moisture. The Jena Experiment has been funded by
The role of plant species richness, plant functional the Deutsche Forschungsgemeinschaft (DFG, Wi 1601/4-1,2,
richness and plant biomass in relation to functional FOR 456). MR acknowledges funding by the German Federal
group composition Foundation for Environment (DBU). We are very grateful to
suggestions from J. Engel, M. F. Gizzie-Voigt and C. Mulder
which substantially improved this manuscript.
Plant species richness and plant functional richness are
weaker predictors of consumer functional group composi-
tion in relation to plant composition. Legume composition
References
alone explains more variation in the composition of each
functional group than species richness or functional richness Alberti, J. et al. 2010. Abiotic stress mediates topdown and
(Fig. 4). Also taking into account the amount of explained bottomup control in a southwestern Atlantic salt marsh.
variance of the remaining plant functional groups, it indicates Oecologia 163: 181191.

975
Aquilino, K. M. et al. 2005. Reciprocal eects of host plant and topdown forces in natural communities. Ecology 73:
natural enemy diversity on herbivore suppression: an empirical 724732.
study of a model tritrophic system. Oikos 108: 275282. Kagata, H. and Ohgushi, T. 2006. Bottomup trophic cascades
Beals, M. L. 2006. Understanding community structure: a data- and material transfer in terrestrial food webs. Ecol. Res.
driven multivariate approach. Oecologia 150: 484495. 21: 2634.
Blanchet, F. G. et al. 2008. Forward selection of explanatory Koricheva, J. et al. 2000. Numerical responses of dierent trophic
variables. Ecology 89: 26232632. groups of invertebrates to manipulations of plant diversity
Blaum, N. et al. 2011. How functional is functional? Ecological in grasslands. Oecologia 125: 271282.
groupings in terrestrial animal ecology: towards an animal Kos, M. et al. 2011. Relative importance of plant-mediated
functional type approach. Biodiv. Conserv. 20: 23332345. bottomup and topdown forces on herbivore abundance
Bommarco, R. and Banks, J. E. 2003. Scale as modier in vegeta- on Brassica oleracea. Funct. Ecol. 25: 11131124.
tion diversity experiments: eects on herbivores and predators. Legendre, P. and Gallagher, E. D. 2001. Ecologically meaningful
Oikos 102: 440448. transformations for ordination of species data. Oecologia
Borcard, D. and Legendre, P. 2002. All-scale spatial analysis of 129: 271280.
ecological data by means of principal coordinates of neighbour Legendre, P. et al. 2005. Analyzing beta diversity: partitioning
matrices. Ecol. Modell. 153: 5168. the spatial variation of community composition data. Ecol.
Borcard, D. et al. 2004. Dissecting the spatial structure of eco- Monogr. 75: 435450.
logical data at multiple scales. Ecology 85: 18261832. Moran, M. D. 2003. Arguments for rejecting the sequential
Borer, E. T. et al. 2006. Asymmetry in community regulation: Bonferroni in ecological studies. Oikos 100: 403405.
eects of predators and productivity. Ecology 87: Oelmann, Y. et al. 2011. Plant diversity eects on aboveground
28132820. and belowground N pools in temperate grassland ecosystems:
Boyer, A. G. et al. 2003. Seasonal variation in topdown and development in the rst 5 years after establishment. Global
bottomup processes in a grassland arthropod community. Biogeochem. Cycles: 25: in press
Oecologia 136: 309316. Oksanen, J. et al. 2010. vegan: Community ecology package.
Cardinale, B. J. et al. 2011. The functional role of producer R package ver. 1.17-4.
diversity in ecosystems. Am. J. Bot. 98: 572592. Peres-Neto, P. R. et al. 2006. Variation partitioning of species data
Chen, Y. G. et al. 2010. Eects of nitrogen fertilization on tri- matrices: Estimation and comparison of fractions. Ecology
trophic interactions. ArthropodPlant Interactions 4: 8194. 87: 26142625.
Crist, T. O. et al. 2006. Spatial variation in insect community and Perner, J. et al. 2005. Eects of plant diversity, plant productivity
species responses to habitat loss and plant community compo- and habitat parameters on arthropod abundance in montane
sition. Oecologia 147: 510521. European grasslands. Ecography 28: 429442.
Dray, S. et al. 2009. packfor: Forward selection with permutation Randlkofer, B. et al. 2010. Vegetation complexity the inuence
(Canoco p.46). R package ver. 0.0-7/r58. of plant species diversity and plant structures on plant chemical
Dyer, L. A. and Letourneau, D. 2003. Topdown and bottomup complexity and arthropods. Basic Appl. Ecol. 11: 383395.
diversity cascades in detrital vs living food webs. Ecol. Lett. Roscher, C. et al. 2004. The role of biodiversity for element
6: 6068. cycling and trophic interactions: an experimental approach
Graham, M. H. 2003. Confronting multicollinearity in ecological in a grassland community. Basic Appl. Ecol. 5: 107121.
multiple regression. Ecology 84: 28092815. Rzanny, M. and Voigt, W. 2012. Complexity of multitrophic inter-
Gripenberg, S. and Roslin, T. 2007. Up or down in space? actions in a grassland ecosystem depends on plant species
Uniting the bottomup versus topdown paradigm and spatial diversity. J. Anim. Ecol. 81: 614627.
ecology. Oikos 116: 181188. Schaers, A. P. et al. 2008. Arthropod assemblages are best pre-
Gruner, D. S. 2004. Attenuation of topdown and bottom dicted by plant species composition. Ecology 89: 782794.
up forces in a complex terrestrial community. Ecology Scherber, C. et al. 2006. Eects of plant diversity on invertebrate
85: 30103022. herbivory in experimental grassland. Oecologia 147: 489500.
Haddad, N. M. et al. 2001. Contrasting eects of plant richness Scherber, C. et al. 2010. Bottomup eects of plant diversity on
and composition on insect communities: a eld experiment. multitrophic interactions in a biodiversity experiment.
Am. Nat. 158: 1735. Nature 468: 553556.
Haddad, N. M. et al. 2009. Plant species loss decreases arthropod Spehn, E. M. et al. 2002. The role of legumes as a component of
diversity and shifts trophic structure. Ecol. Lett. 12: biodiversity in a cross-European study of grassland biomass
10291039. nitrogen. Oikos 98: 205218.
Hairston, N. G. et al. 1960. Community structure, population Vehvilinen, H. et al. 2008. Eects of stand tree species compo-
control and competition. Am. Nat. 94: 421425. sition and diversity on abundance of predatory arthropods.
Halaj, J. and Wise, D. H. 2001. Terrestrial trophic cascades: Oikos 117: 935943.
how much do they trickle? Am. Nat. 157: 262281. Voigt, W. et al. 2007. Using functional groups to investigate com-
Halpern, B. S. et al. 2006. Strong topdown control in southern munity response to environmental changes: two grassland case
California kelp forest ecosystems. Science 312: 12301232. studies. Global Change Biol. 13: 17101721.
Hoekman, D. 2007. Topdown and bottomup regulation in a Weigelt, A. et al. 2010. The Jena experiment: six years of data from
detritus-based aquatic food web: a repeated eld experiment a grassland biodiversity experiment. Ecology 91: 929.
using the pitcher plant (Sarracenia purpurea) inquiline com- White, T. C. R. 1978. The importance of a relative shortage of
munity. Am. Midl. Nat. 157: 5262. food in animal ecology. Oecologia 33: 7186.
Hoekman, D. 2010. Turning up the heat: temperature inuences Wimp, G. M. et al. 2010. Increased primary production shifts
the relative importance of topdown and bottomup eects. the structure and composition of a terrestrial arthropod
Ecology 91: 28192825. community. Ecology 91: 33033311.
Hunter, M. D. and Price, P. W. 1992. Playing chutes and ladders Zhang, Y. J. and Adams, J. 2011. Topdown control of herbivores
heterogeneity and the relative roles of bottomup and varies with ecosystem types. J. Ecol. 99: 370372.

Supplementary material (available as Appendix oik-00114


at  www.oikosoce.lu.se/appendix ). Appendix 12.

976

You might also like