Linking traits between plants and invertebrate herbivores to track functional effects of land-use changes

Questions: Ecosystem functions and underlying services are strongly influenced by multitrophic relationships, with functional traits playing a central role in structuring them. Which traits and functional metrics mediate the impact of different types of land use on ecosystem function within and across trophic levels? Methods: We studied the functional relationships between plants and grasshoppers in sub‐alpine grasslands under different management regimes in the Central French Alps. We applied the theoretical multitrophic response–effect framework described by (Journal of Vegetation Science, 24, this issue) to identify key traits linking plants and grasshoppers to biomass production. The linkages between selected plant and grasshopper traits were analysed using community‐weighted mean traits (CWM) and functional diversity (FD; Rao's quadratic diversity). Results: Uni‐ and multivariate models provided evidence about the relative importance of trait linkages within and across trophic levels. We showed that management affected both plant and grasshopper traits and that the interaction between them was linked to biomass production. While a number of CWM traits and FD were involved in the interaction, CWM of leaf dry matter content (LDMC) and grasshopper dry body mass (GMass) chiefly mediated the impact of management change on biomass production. Conclusions: Our study suggests that both trait values of the most abundant species and functional trait variation within and across trophic levels in combination may best explain the impact of land‐use changes on ecosystem function. To improve our mechanistic understanding across trophic levels, a better knowledge of response and effect traits remains a major goal, especially for animal ecologists, while a strong collaboration among disciplines is needed to bridge the existing gaps.

Ecosystem functions and underlying services are strongly influenced by multitrophic relationships, with links between response and effect traits playing a central role. We studied the functional relationships between plant and grasshopper communities and found that both dominant traits in the community and trait variation within and across trophic levels drive ecosystem functions.

CWM; Ecosystem functioning; Ecosystem services; Functional dissimilarity FD; Functional traits; Herbivory; Multitrophic trait cascade; Plant–herbivore interaction

Nomenclature

http://www.faunaeur.org/

Predicting how global changes will affect ecosystem functions and resultant services through changes in biological communities is one of the most urgent tasks in ecology. Recent syntheses and empirical studies have highlighted that functional traits (sensu Violle et al. [77] ) better predict the effects of global changes on ecosystem services than taxonomic parameters alone (Hooper et al. [33] ; Díaz et al. [20] ,[21] ; de Bello et al. [4] ; Lavorel et al. [43] ).

Many ecosystem functions ultimately rely on interactions between primary producers and other trophic levels, such as pollinators, soil decomposers and herbivores (Loreau et al. [48] ; Kremen et al. [39] ). For example, nitrogen cycling involves complex interactions between plants and soil biota (Hyvonen et al. [34] ), and between plants and herbivores (Heidorn & Joern [31] ; Bardgett & Wardle [1] ; Dubey et al. [23] ). However the mechanisms that drive the relationships between biodiversity and ecosystem functioning are still poorly known despite growing evidence for the importance of trophic interactions (Thébault & Loreau [73] ; Suding & Goldstein [72] ). Up to now only a few studies have considered multitrophic systems (Downing & Leibold [22] ; Fox [25] ; Duffy et al. [24] ; Ives et al. [37] ; Thébault & Loreau [73] ), and to our knowledge they considered species diversity rather than functional components of biodiversity.

There is a growing hope that extending the trait concept to multitrophic systems will improve our ability to identifying mechanisms that drive biotic control over ecosystem service delivery (e.g. Schmitz [65] ; de Bello et al. [3] ,[4] ; Ibanez [35] ). Most studies of the trophic interactions have been focused at the species level, addressing the question ‘who eats whom’ and ‘how much’. Few attempts have been proposed to quantify relationships across trophic levels using functional traits of single species, while no attempt has been made to link functional traits at community level across trophic guilds (Stang et al. [71] ; de Bello et al. [3] ; Brooks et al. [14] ; Ibanez [35] ). In this paper we apply the multitrophic response–effect traits framework proposed by Lavorel et al. (in press) to analyse functional relations that underlie responses to exogenous drivers and effects on ecosystem functioning. The framework is based on response–effect traits theory (Lavorel & Garnier [40] ; Violle et al. [77] ; Suding & Goldstein [72] ), and further considering traits responsible for trophic interactions. The framework therefore considers three types of traits: (1) morphological, physiological and behavioural traits related to fitness and ability of organisms to cope with different environmental conditions (i.e. ‘response traits’); (2) ‘trophic traits’ that are involved in the interaction across trophic levels (e.g. nutritional characteristics, palatability and toughness on one side, and size, shape and strength of the consumption organs on the other), and finally (3) traits involved in ecosystem effects, i.e. ‘effect traits’ (e.g. consumption rate of herbivores or decomposers). The problem is that often such traits are not yet known, so proxies (i.e. soft traits – sensu Hodgson et al. [32] ), often from the literature, are used with a risk of losing explanatory power in the model.

Two widely used distinct metrics of community trait composition are hypothesized to affect ecosystem processes: (1) the dominant trait value in the community, quantified by the community‐weighted mean of trait values (CWM), and (2) the degree to which trait values differ in a community, quantified by functional diversity (FD) (see Ricotta & Moretti [63] for a synthesis). CWM effects translate the mass ratio hypothesis (Grime [30] ), which proposes that ecosystem processes are mainly determined by the trait values of the dominant species present in a community, hence, that a species’ influence on ecosystems should be proportional to its relative contribution to the total number or biomass of the community (Garnier et al. [28] ; Quested et al. [59] ). On the other hand, FD is associated with the complementary resource use hypothesis (Tilman et al. [74] ), and proposes that when there is higher FD among the species present in a community, reflecting higher diversity of resource use strategies, there will be a more complete exploitation of resources than in less functionally diverse communities (Petchey et al. [58] ; Hooper et al. [33] ; Spehn et al. [70] ). It still remains basically unexplored as to whether multitrophic trait linkages are affected by these community parameters (Reiss et al. [62] ).

In this paper we use plants and grasshoppers as model organisms to assess the links between their traits and to quantify the relative importance of environmental and trophic factors to explain species and functional traits composition. We used the term ‘grasshoppers’ for simplicity, but we considered both true grasshoppers (Caelifera) and bush‐crickets (Ensifera). These groups can have a relatively high impact in alpine grasslands, consuming up to 30% of the green biomass (Blumer & Diemer [8] ). They display great inter‐specific differences in mandibles and head morphology, leading to distinct food preferences and diets (e.g. Bernays et al. [6] ). Here, we aimed to identify (1) which trait metrics respond to different management types; (2) which trait metrics relate to biomass production; and (3) which trait metrics link the response to effect traits within and across trophic levels. With this last question, which to our knowledge has not yet been addressed at the community level under natural conditions, we propose to (1) advance the fundamental understanding of mechanisms underlying relationships between biodiversity and ecosystem functioning by considering FD at multiple trophic levels, and (2) provide a first attempt to identify key traits and metrics that link effects of environmental changes with the ecosystem processes and ecosystem services delivered by biodiversity in a multitrophic context.

Based on existing knowledge of plant and grasshopper trait responses to changing grassland management regime, their potential trophic linkage and effects on the biomass production, we expect that fertilization favours plants with higher specific leaf area, higher concentrations of leaf nutrients such as nitrogen or phosphorus and lower dry matter content (e.g. Quétier et al. [60] ). Such plants would provide highly palatable food for grasshoppers (Pérez‐Harguindeguy et al. [56] ) that are large, have little mobility and live in dense vegetation. Grazing is expected to favour well‐protected tussocks of small and rather unpalatable plants with lower nitrogen content (Díaz et al. [20] ,[21] ; Loranger et al. in press) and small tough leaves (Pérez‐Harguindeguy et al. [57] ; Peeters et al. [55] ). Such plant traits are likely to sustain grasshopper species with a large head (Bernays & Hamai [5] ) and mandibles (Isely [36] ), while influencing oviposition site substrate due to the overall change in the canopy structure of the herbaceous layer, with both patches of bare soil among grass clumps due to trampling effect and accumulated unpalatable plant litter (O'Neill et al. [53] ). Based on Specht et al. ([69] ) and Unsicker et al. ([75] ), we expect plant diversity and plant functional identity, i.e. the variation rather than the dominance of plant traits, to positively affect grasshoppers’ fitness and traits related to potential fecundity (e.g. number of ovarioles). Concerning repercussions of vegetation and grasshopper changes on primary productivity, we expect low leaf nitrogen content and high leaf tissue density to reduce both fodder productivity and quality of plant communities (Pérez‐Harguindeguy et al. [56] ). Lower leaf palatability would then negatively impact herbivory by grasshoppers (Bownes et al. [11] ), slowing down ecosystem‐level nutrient cycling, and leading to lower productivity in grazed compared to mown and fertilized grasslands (Lavorel & Garnier [40] ; Quétier et al. [60] ). Finally, based on the most recent investigations, we expect the combination of dominant traits (CWM) and variation in traits (FD) to drive ecosystem processes (e.g. Mokany et al. [51] ; Schumacher & Roscher [66] ; Mouillot et al. [52] ), although very little is known on the relative importance of the two functional metrics across trophic levels. We are aware that using observational studies can leave uncertainties as to the causal relationships between trait metrics and trophic levels and that our study presents mainly co‐occurrence rather than strict trophic relationships. Nevertheless, we believe that our study represents a valuable insight into trait linkages within and across trophic levels.

The study area is located in the upper valley of the Romanche River, central French Alps (Villar d'Arène, 45°2′24″ N, 6°20′24″ E). The substrate is homogenous calc‐shale and the climate is sub‐alpine with a strong continental influence. The landscape is dominated by sub‐alpine meadows and pastures with distinct past and current management regimes. At the lower altitudes (1650–2000 m), former arable fields have been abandoned and subsequently converted to terraced grasslands used for hay or grazing. At mid‐slope (1800–2100 m), ancient, never‐ploughed hay meadows are increasingly left for light summer grazing by sheep or cattle. The study area is part of a larger experimental area with several management trajectories and experimental plots representing past and current management regimes. For more details see Quétier et al. ([60] ) and Lavorel et al. ([43] ).

We selected five management trajectories (sensu Quétier et al. [60] ) representing the main management practices that best reflected the past and current management of these sub‐alpine grasslands (Table [NaN] ). In this system, any quantitative variable such as disturbance regime or soil fertility only partially captures management effects by ignoring past land‐use legacies (Quétier et al. [60] ). In each trajectory we selected three sampling plots (replicates), of 200 m2 each, at different locations and altitudes from 1650 to 1900 m a.s.l. to avoid spatial autocorrelation. Hereafter, we will use the term ‘management’ when we refer to the five trajectories, and ‘plots’ when we refer to sampling plots.

Past and current management of the five selected management trajectories. Past management indicates the main practice in the last century until the most recent change, which determined the current use (for more details, see Quétier et al. 60 )

Floristic composition of each plot was assessed using three non‐intersecting point‐quadrat survey lines (Daget & Poissonet [17] ). These were 8‐m long, with one point every 20 cm for a total of 120 points per plot. The total number of contacts for each species was counted at each point of the point‐quadrat survey. Species accounting for 80% of total accumulated abundance were selected for trait measurements. Based on current knowledge (see Introduction) we selected traits (Table [NaN] a) expected to respond to management and to affect both biomass production (i.e. annual net primary production) and grasshopper traits (mainly through nutritional properties, palatability, but also as source of habitat structures and habitat conditions). We selected six traits: plant vegetative height (VegH), leaf tensile strength (leaf toughness, Tough), leaf dry matter content (LDMC), specific leaf area (SLA), leaf nitrogen content (LNC) and leaf phosphorus content (LPC) (Quétier et al. [60] ; Lavorel et al. [43] ). The detailed trait measurements followed standard protocols (Cornelissen et al. [16] ; Quétier et al. [60] ). Plant traits were measured on ten randomly selected plant individuals per species and per trajectory in order to capture intra‐specific variation in response to management (Garnier et al. [29] ). Primary production was assessed at peak biomass date (from early to mid‐July, depending on trajectories) using calibrated visual estimates over 12 quadrats per plot (see Lavorel et al. [42] ; Redjadj et al. [61] ).

Plant (a) and grasshopper (b) traits used in the analyses (*indicates that the trait was measured on at least ten individuals)

1 Plant traits were measured by Quétier et al. (60) following Cornelissen et al. (16).

• 2 Organs containing the eggs in insects and spiders. In insects they are estimators of potential.
• 3 fecundity.
• 4 See details in Appendix S1.

Grasshoppers were sampled using a biocenometer, a quantitative box sampling method (Gardiner & Hill [27] ) that enables the estimation of grasshopper population densities and community structures. The biocenometer consists of a 1 × 1 × 1 m cube made of tissue and open at the bottom (to be placed on the ground) and top (to collect the animals by hand). A 2‐m long aluminium support allowed operators to lift, move and place the biocenometer at the different sampling sites without disturbing the grasshoppers. Each time the biocenometer was placed firmly on the ground and the enclosed 1 m2 was examined for grasshoppers. All collected grasshoppers were kept in alcohol in separate vials (one per plot) and transported to the lab. All adults and sub‐adults were identified at species level using standard keys (Coray & Thorens [15] ; Bellman & Luquet [2] ). At each vegetation plot, we sampled grasshoppers 15 times, i.e. five samples along three parallel transects (5‐m distance between samples). The sampling was repeated at three distinct times during the vegetation and grasshopper season, i.e. at the end of July, at the end of August and at the end of September 2008. Samples of each plot were pooled.

We selected grasshopper traits (Table [NaN] b) expected to respond to management (mainly mowing, grazing and fertilization). These are traits sensitive to habitat structure and climate conditions, such as body size, dry body mass and oviposition site (on the ground or in the vegetation), traits enabling response to disturbance, such as mobility (ability to escape or recolonize disturbed surfaces), and traits expected to be linked to plant traits as habitat and food sources (e.g. diet mainly based on grass or forb, mandible size and body size). Some of these traits are expected to respond to both management and plant traits, and to have a link to primary production, mainly through consumption (e.g. body size, dry body mass, mandible size; see Introduction). Trait data was sourced from literature information (i.e. Detzel [18] ) and complemented by morphological measurements taken on ten individuals (five males and five females) of each adult species, randomly sampled among the five trajectories (see Appendix S1).

For each plant and grasshopper trait we calculated two functional metrics widely used in functional community studies (e.g. Díaz et al. [20] ,[21] ; Vandewalle et al. [76] ). First, the community‐weighted mean (CWM) trait value (Garnier et al. [28] ), which expresses the mean trait value in the community weighted by the relative abundance of the species. CWM reflects the average trait value of the most dominant species in a community, which has been interpreted as translating the mass‐ratio hypothesis by (Grime [30] ), i.e. the dominant traits in a community exert the greatest effect on ecosystem functions. Second, Rao's quadratic entropy of functional diversity (FD), which equals the sum of the dissimilarity in trait space among all possible pairs of species, weighted by the product of the species’ relative abundance (Botta‐Dukat [10] ; Lepš et al. [46] ; Ricotta & Szeidl [64] ; de Bello et al. [3] ). Details of the formulas can be found in Ricotta & Moretti ([63] ). For plants and grasshoppers, mostly uncorrelated functional metrics for each trait (< 0.5 Pearson correlation) were selected. Traits providing distinct and complementary ecological and functional information were included in the analyses. In the text, the functional metric of the different traits appears beside the name of each trait (TraitCWM; TraitFD).

Data analyses were designed to identify the traits involved in the multitrophic response–effect traits framework (as shown in Fig. [NaN] a) and further described by Lavorel et al. (in press). We applied a multistep analysis using a variety of uni‐ and multivariate linear regressions. Models were based on both full and partial analyses (using covariables) on different sets of variables to find relationships between land use, plant and grasshopper traits and their relative contributions to biomass production. Figure [NaN] b shows an overview of the different data sets used and the analytical steps performed with respect to the multitrophic response–effect traits framework (Fig. [NaN] a).

In a first step, we identified the plant (step 1a in Fig. [NaN] ) and grasshopper (step 1b) response traits (DR1 and DR2, respectively, in Fig. [NaN] a) that were significantly affected by management (i.e. the five trajectories) used as explanatory variables in ANOVA (see Fig. [NaN] b). In step 2a, we identified single grasshopper traits (TR2) responding to plant traits (TE1) in Fig. [NaN] a. To do this, we first ran a forward selection with the R package packfor (R Foundation for Statistical Computing, Vienna, AT) using redundancy analyses (RDA) with plant traits as explanatory variables and grasshopper traits as response variables to identify, among all plant traits, those trait metrics affecting the ensemble of grasshopper traits (P < 0.05 after 9999 random permutations). We minimized problems of classical forward selection by applying the double‐step procedure proposed by Blanchet et al. ([7] ): (1) inflated Type I error was avoided by forward selecting only models for which a global test with all explanatory variables was significant; (2) to avoid overestimation of the amount of variance explained, an additional stop criterion was introduced, in that the adjusted coefficient of multiple determination (R2 adj) of the model could not exceed the R2 adj obtained when using all explanatory variables. The variables that fulfilled both stop criteria were identified as significant environmental variables shaping the communities of the focal taxa.

Being aware of the limits of any selection procedure, we ran exploratory regressions on several variable combinations, yielding the same qualitative results (not shown). We then tested the effect of significant selected plant traits on each single grasshopper trait metric, again using RDA, to identify which specific grasshopper parameters responded specifically to vegetation. The same procedure was applied in step 2b (Fig. [NaN] ) to identify possible links of grasshopper traits (TR1) to plant traits (TE2) in Fig. [NaN] a. In step 3 we finally assessed the link between plant/grasshopper traits and biomass production (i.e. FE1 and FE2, respectively, in Fig. [NaN] a), using the standing dry biomass of the vegetation as response variable. The plant and grasshopper traits that were identified as responding to management and/or the other trophic level and traits affecting biomass production represented the linkage between response and effect traits (step 4).

Finally, variation partitioning (Borcard & Legendre [9] ; Legendre et al. [45] ) was used to quantify the relative importance of the variables (management and traits) involved in the different steps. By using management as a covariable, we intended to remove the variance explained by the main environmental factors that may confound the trophic links within and across trophic levels. Based on these results, we tested the paths connecting management and key plant/grasshopper traits to biomass production, using path analysis, which is a suitable method to describe the direct dependencies among a set of variables (Shipley [67] ).

The statistical analyses were performed in R version 2.13.1 using the library ‘vegan’ 1.17–3, ‘sem’ 0.9–21 and ‘packfor’.

In the next four sections we first present the results for each single step (1–4) used in the analyses to untangle the different components of the trait framework across trophic levels (Fig. [NaN] ). At each step we quantify the unique and shared variance explained by the different sets variables (management and traits) involved (results synthesized in Fig. [NaN] ). Finally, we assessed the links between the traits responding to management (environment response traits) and the other traits across trophic levels (trophic response–effect traits linkage) with the traits affecting ecosystem functions (ecosystem effect traits). Appendix S2 provides Pearson correlation values and sign of the correlation between single plant and grasshopper traits and biomass production. This is useful to understand the direction of the linkages between variables to interpret the multitrophic framework (Fig. [NaN] ) and path analysis (Fig. [NaN] ).

Overall, we observed 501 grasshopper individuals from 18 species. Of these, four species were dominant (Stauroderus scalaris, Omocestus haemorrhoidalis, Chorthippus apricarius and Euthystira brachyptera) with relative abundances > 10% in each of the 15 plots and 49% of the total number of individuals (Appendix S3). For the vegetation, patterns have already been described in Quétier et al. ([60] ).

Management significantly affected both plant and grasshopper traits (Table [NaN] column M, i.e. management). The CWM of all six plant traits (Table [NaN] a) was significantly affected, with fertilization and mowing (Trajectory 1) positively affecting LNCCWM, LPCCWM and SLACWM; vice versa for LDMCCWM in Trajectory 1 and, partially, in 5 (lightly grazed hay meadow). The latter, together with mowing without fertilization (Trajectory 4), positively affected toughness (ToughCWM; data no shown). FD responded only for leaf toughness (ToughFD) and vegetation height (VegHFD), with a positive effect in all management types except strong grazing.

(a,b) P‐values of the uni‐ and multivariate analyses of full and partial models of the different explaining set of variables linking traits across trophic levels

5 (Significant values are in bold; ‘–’ not significant or not selected). M, management types; Gt, grasshoppers traits; Pt, plant traits; B, biomass production. Numbers and letters indicate steps of the analyses shown in Figs 1b and 2 (the symbol ‘|’ means that the effect of the following variable was removed (partial model), e.g. M|Gt: effect of M after removal of Gt variance. Trait names are given in Table .

Grasshopper traits responded to management in more diverse ways. CWM of number of ovarioles (NOvarCWM) and FD of mandible width (MandWFD) were higher in fertilized and mown hay meadows (Trajectory 1), while mobility (HMobilCWM) and mandible width (MandWCWM) were lower. Mobility was also low in lightly grazed meadows, while mandible width was high.

Overall we observed that management alone explained 86.5% of the variance in plant traits, while it accounted for only 8.4% of the grasshopper traits, but 52.6% when in combination with plant traits (Fig. [NaN] ; results of variance partitioning using full and partial models; see Methods). Plant traits alone accounted for 29.3% of the variance of grasshopper traits and this increased to 83.5% when combined with management. Thus a large variation in grasshopper traits seems to depend on the effects of vegetation, directly or via changes in management.

While plant traits affected grasshopper traits (step 2a in Fig. [NaN] , see also above), no grasshopper trait showed a clear effect on plant traits (step 2b). Forward selection highlighted four plant traits (LDMCCWM,FD, LNCCWM, LPCFD, SLAFD) as significantly (or marginally) linked to grasshopper traits (Table [NaN] a column Gt). These four plant traits particularly effected several grasshopper traits, i.e. CWM of number of ovarioles (NOvarCWM), mobility (HMobilCWM), body length (BodyLCWM), dry body mass (GMassCWM) and FD of the two latter traits (GMassFD; BodyLFD) and tibia length (TibiaLFD) (see Table [NaN] b column Pt and Appendix S2 for direction of the correlations). Four traits were still significantly affected by the four plant traits above after removing the effect of management (see Table [NaN] b column Pt|M). Overall, NOvarCWM and HMobilCWM were the only two grasshopper traits affected by both plant traits and management (Fig. [NaN] ).

Biomass production was negatively affected by CWM of both leaf dry matter content (LDMCCWM) and grasshopper dry body mass (GMassCWM) (forward selection of plant and grasshopper traits used separately as explanatory variables and biomass production as response variable; Table [NaN] column B). Most of the linkages, however, could be explained by the combination of plant and grasshopper traits when structured by management (61.2%, see Fig. [NaN] c, i.e. 0.214 + 0.227 + 0.030 + 0.140 + 0.001). The contribution of management to the overall variance of biomass production was 54.6% (i.e. 0.075 + 0.214 + 0.227 + 0.030) mainly structuring plant effect traits (44.1% shared variance, i.e. 0.214 + 0.227) vs. 25.7% (i.e. 0.227 + 0.030) shared with grasshoppers. Unique contributions to biomass production by management only (7.5%), vegetation traits only (0%) and grasshopper traits only (14%) were comparably low.

Two community mean plant traits, LDMCCWM, FD and LNCCWM, were affected by management and linked to grasshopper traits: (Table [NaN] a column Gt[2a] and Table [NaN] b col. Pt[2a]). Two additional trait metrics depicting functional divergence within plant communities, LPCFD and SLAFD, were only linked to grasshopper traits, while LDMCCWM was also linked to biomass production (Table [NaN] a col. B and Fig. [NaN] rows 2b and 3a). Thus the direct effect of plant traits on grasshoppers and biomass production appeared to be mediated by LDMCCWM. For grasshoppers, CWM of dry body mass (GMassCWM) responded to plant traits and was linked to biomass production (Fig. [NaN] rows 2b and 3b). No single grasshopper metric connected trait responses to management to biomass production. In other words, the connection between management and biomass production via grasshoppers stemmed from plant community properties, LDMCCWM,FD and LNCCWM affecting GMassCWM. Thus, LDMCCWM and GMassCWM appeared as the key traits in linking management to biomass production across trophic levels.

The general patterns that emerged from our data analysis were confirmed when we tested the hypothesized links between management, LDMCCWM, GMassCWM and biomass production using path analysis and structural equation modeling/SEM (Model Chi squared = 0.475, df = 1, P = 0.491; note P > 0.05 indicates significance for SEM; see Fig. [NaN] ). While significant pathways were found from management to LDMCCWM (standardized regression path coefficients R2 0.75) and from LDMCCWM to biomass production (R2 −0.58), the negative link between GMassCWM and biomass production (R2 −0.47) was mediated by a positive significant trophic link between LDMCCWM and GMassCWM (R2 0.40). The direct pathway from management to biomass production (R2 0.35) was, instead, not significant.

Our study represents one of the first attempts to apply the trait framework proposed by Lavorel et al. (in press), although we are aware that others are working on similar topics. Our analyses clearly demonstrate the usefulness of the approach using plant and grasshopper communities under natural conditions. Our analytical approach allowed us to tease out the relative contributions of the several components involved in the multitrophic trait framework. It also provided insights into possible key trait‐based mechanisms that underlie ecosystem services responses to environmental change. We are aware that such correlative relationships cannot necessarily be interpreted as cause–effect, but rather simple links between variables. Using partial uni‐ and multivariate models with management as a covariable should, however, have removed most of the variation due to confounding environmental factors, at least those that are structured by management. Our results highlight two major aspects, which we discuss below: first, the trait identity affecting biomass production, and second, the relative role of community‐weighted mean traits (CWM) and functional divergence (FD).

Our study is the first, as far as we are aware, to show the effect of an environmental driver (management) on an ecosystem function (biomass production) mediated by response‐and‐effect traits across two trophic levels, i.e. plants and grasshoppers. Taxonomic and functional responses of plants and grasshoppers to management in meadow ecosystems are relatively well known. In our study, response traits of both trophic levels were consistent with other studies and with our expectations (see Introduction for an overview). Only two plant traits responding to management were further involved in either biomass production (leaf dry matter content. LDMCCWM) or in the trophic linkage (leaf nitrogen content, LNCCWM and LDMCCWM,FD).

In particular, plant species with high leaf dry matter content (LDMCCWM) were mainly favoured in unfertilized meadows (Trajectory 4) and in lightly grazed pastures (Trajectory 5), and played a key role in the trait cascade. This trait linked the response of plant community composition to management and biomass production both directly as an effect trait and indirectly as a trophic effect trait on grasshoppers, and especially on the dominance of species with high dry body mass (GMassCWM). Grasshopper dry body mass indeed acted as an effect trait on primary production, with a negative link to plant biomass. The reasons for the positive link between LDMCCWM and GMassCWM could be trophic, making, e.g. hard leaves more palatable and assimilated by grasshoppers, but also related to habitat conditions and vegetation structure. Underlying mechanisms explaining the links between leaf chemical and structural properties and grasshopper body mass and related traits (e.g. mandible width) remain to be tested with appropriate experiments to remove the influence of biotic and abiotic habitat constraints, such as microhabitat, climate, plant architecture and predators (e.g. Joern & Lawlor [38] ; Marini et al. [49] ; Gardiner & Hassal [26] ). Biomechanical properties of the mandibles and other mouthparts (Patterson [54] ) or physiological and morphological adaptations of the gastrointestinal tract (Martin et al. [50] ) may play an important role. Furthermore, several studies have confirmed the relevance of nutrient acquisition/conservation trade‐off governing plant nutrient economy (Díaz et al. [19] ; Wright et al. [78] ; Quétier et al. [60] ; Lavorel & Grigulis [41] ) to predict responses of plant communities to decreasing management intensity. Here, we demonstrated explicitly how such trade‐offs could have important consequences on other trophic levels and on ecosystem functioning (see Lavorel & Grigulis [41] ). Indeed, grasshoppers are highly sensitive to the protein:carbohydrate ratio, a trait governed by the nutrient acquisition–conservation trade‐off (Simpson & Raubenheimer [68] ).

Management also largely influenced dominant functional characteristics of grasshoppers related to feeding capacity and variation (MandWCWM,FD), dispersal (HMobilCWM) and reproduction (NOvarCWM), with mown and fertilized grasslands (Trajectory 1) favouring communities dominated by species with low mobility (indicating rather stable conditions), as well as small and diversified mandibles (indicating availability of different trophic niches) that have a high number of ovarioles (i.e. high potential fecundity). Only mobility and number of ovarioles were involved in the plant–grasshopper linkage, with high quality food (high LNC and LPC) potentially favouring fecundity of grasshoppers (e.g. Branson [12] ). Nevertheless, none of these traits was further linked to primary production, which does not exclude a possible overlap between grasshopper response and effect traits on other ecosystem processes not considered in our study. No linkages were found between grasshopper and plant traits (step 2b). The strong impact of management on plant traits found in our results could have masked possible top‐down control of plant traits by grasshoppers. At the same time, we believe that the missing link between grasshoppers and plants could also add new insights into the effect of herbivory on plant composition when functional traits rather than species are considered (see Branson & Sword [13] ).

Only 3% of the overall variation of biomass production was explained by the combination of management and grasshopper effect traits, while 22.7% was mediated through the linkage between plant and grasshopper effect traits. This confirms the important role of the trait link across trophic levels mediating ecosystem function.

Our study revealed two main aspects related to the different steps in the multitrophic response–effect trait framework. First, we found that community‐weighted means (CWM) were the main driving functional metrics relevant to the two key response–effect traits (LDMC and GMass) in the multitrophic trait framework, highlighting the important role of the dominant trait values in both communities, rather than the variance in trait composition (FD). CWM was also the major overlapping functional metric between management response traits and functional effect traits affecting biomass production in both plants and grasshoppers. This finding confirmed the prediction of the biomass ratio hypothesis (Grime [30] ).

Second, trait variation within communities of plants and grasshoppers (trait dissimilarity, FD) was also important in the link across trophic levels, but it did not have a direct or indirect effect on biomass primary production. Plant species diversity (Unsicker et al. [75] ) and plant functional identity (Specht et al. [69] ) have been demonstrated to increase the performance and the fitness of generalist grasshoppers. Both resource dominance and variability thus appear to be important in plant communities for the trophic linkage. Nevertheless, the combined presence of both functional metrics (CWM and FD) in the trophic linkage between plants and grasshoppers might reveal a confounding effect by several components governing trophic interactions in the field, e.g. ‘habitat’ (i.e. plant community structuring of the environment occupied by grasshoppers), ‘food availability’ (i.e. the same plant community in the habitat also representing food sources) and ‘food preferences’ (i.e. the plant species selected by a grasshopper based on nutritional and biomechanical properties). Since plant composition in the sampling plots represented both the habitat and food resources for the grasshoppers, we cannot untangle these two components to address the mechanism(s) underlying the effect–response trait linkages between plants and grasshoppers. Trait dissimilarity (FD) of plant communities might be important to provide a sufficient variety of food sources in a diversified habitat satisfying specific abiotic requirements (Unsicker et al. [75] ), reflecting niche partitioning and relative resource exploitation, while dominant plant traits (CWM) might drive food selection by the different grasshopper species, satisfying precise nutritional needs and biomechanical constraints.

The complementary role of CWM and FD in ecosystem processes, such as primary production, in natural systems is consistent with recent evidence (e.g. Mokany et al. [51] ; Schumacher & Roscher [66] ; Mouillot et al. [52] ) suggesting that both trait values of the most abundant species and functional trait variation within and across trophic levels in combination may best explain the impact of environmental changes on ecosystem function.

Overall our results (1) provided strong evidence highlighting the functional links between trophic levels via community trait metrics described by Lavorel et al. (in press); (2) showed the contribution of each partial step in the trait framework; and (3) gave an insight into key trait linkages across trophic levels and their link to biomass production. Our study suggests that while changes in community dominance hierarchies deserve most attention when managing communities for the maintenance of ecosystem processes, communities with diverse trait values should be maintained to optimize trophic interactions.

There is an urgent need to quantify traits for multiple trophic guilds interacting with plants. Particular effort should be devoted to defining and measuring trophic and ecosystem function effect traits. Concerning terrestrial arthropods, most of the traits available in the literature are response traits, which do not directly affect ecosystem functions. We hope that this contribution will stimulate ecologists and statisticians to collaborate to explain the multitrophic interaction webs and the mechanisms mediated by functional traits.

This research was conducted on the long‐term research site Zone Atelier Alpes, a member of the ILTER‐Europe network (ZAA publication no. 23). We received logistic and infrastructure support from the Station Alpine Joseph Fourier (UMS 3370 CNRS‐Université Joseph Fourier) and funding from EraNet BiodivERsA project VITAL (ANR‐08‐BDVA‐008). The Natural History Museum of Lugano provided field‐sampling material. We are grateful to Olivier Manneville (Laboratoire d'Ecologie Alpine and Station Alpine Joseph Fourier, Université Joseph Fourier, Grenoble) for useful advice during the field season.

Graph: ( a, b ) Theoretical multitrophic response–effect trait framework of Lavorel et al. (in press) to link traits, i.e. DR , driver response traits; TE , trophic effect traits; TR , trophic response traits; FE , functional effect traits ( L inkage: DR , TR  =  FE , TE ) across trophic levels 1 and 2. Numbers 1a,b – 4a,b correspond to the different steps (step 1a,b – step 4a,b) in the scheme of the analyses below. ( b ) Diagram of the different data sets used and the analytical steps performed with respect to the theoretical multitrophic response–effect trait framework (see scheme above and Methods). CWM , community‐weighted mean; FD , functional dissimilarity ( R ao's quadratic entropy); dij , dissimilarity between each pair of species i and j ; P t, plant traits; G t, grasshopper traits; [ x ]: single trait; sel : selected traits from the step‐wise procedure (stpw); stpw: forward step‐wise selection of the traits in RDA ; RDA / pRDA : Redundancy analysis/partial RDA ; sqrt‐transf.: square‐root transformation applied on the grasshopper abundance data; ~: linear relationship between response and explanatory variables; |: covariables used in the partial models.

Graph: Synthesis of the results of our study (see also Table  ) fitted into the theoretical framework of Fig.  a, showing variance explained ( R2 adj) of unique and combined fractions of the different variables involved, based on Fig.  (i.e. a 3a; b 3b; c 3c). Full trait names are given in Table  .

Graph: ( a–c ) Variation partitioning with relative variance explained ( R2 adj) of the various unique (outside of the circle interceptions) and combined fractions (inside) of the different explanatory variables of management ( M ), grasshopper traits ( G t) and plant traits ( P t) involved in the trait framework specified by step numbers [1–4, a–b] as in Fig.  a.

Graph: Path diagram showing paths (arrows) connecting management, key plant/grasshopper traits and biomass production. Standardized regression path coefficients are given on the arrows, with the thickness proportional to the path coefficient value. Negative correlations are denoted by dotted and positive by solid lines. P ‐values *< 0.05; **< 0.01; ***< 0.001. Model Chi squared = 0.475, df  = 1, P ‐value = 0.491; Adj. goodness‐of‐fit index = 0.836. For details of the statistics, see Appendix S4.

Graph: Appendix S1. Measurements taken on grasshoppers (dotted line), i.e. (a) body length (BodyL), (b) tibia length (TibiaL) and (c) mandible width (MandW).

Graph: Appendix S2. Pearson correlation matrix between plant and grasshopper traits and between the latter and biomass production. See full trait names in Table . Correlation values > 0.5 are in bold.

Graph

Graph: Appendix S3. List of the grasshopper species sampled in the five land‐use trajectories in the upper valley of the Romanche River (Villar d'Arène) in the central French Alps.

Graph

Graph: Appendix S4. Detailed results of the path analyses in Fig. .

Graph