Herbivore regulation is one of the services provided by plant diversity in terrestrial ecosystems. It has been suggested that tree diversity decreases insect herbivory in forests, but recent studies have reported opposite patterns, indicating that tree diversity can trigger associational resistance or susceptibility. The mechanisms underlying the tree diversity–resistance relationship thus remain a matter of debate.
We assessed insect herbivory on pedunculate oak saplings (Quercus robur) in a large-scale experiment in which we manipulated tree diversity and identity by mixing oaks, birch and pine species.
Tree diversity at the plot scale had no effect on damage due to leaf chewers, but abundance of leaf miners decreased with increasing tree diversity. The magnitude of this associational resistance increased with host dilution, consistent with the ‘resource concentration hypothesis’.
At a smaller scale, we estimated tree apparency as the difference in total height between focal oak saplings and their nearest neighbouring trees. Levels of oak infestation with leaf miners decreased significantly with decreasing tree apparency. As the probability of having taller neighbours increased with tree diversity, notably due to the increase in the proportion of faster growing nonhost trees, such as birches and pines, tree apparency may be seen as a ‘hidden’, sampling effect of tree diversity.
Synthesis. These findings suggest that greater host dilution and lower tree apparency contribute to associational resistance in young trees. They also highlight the importance of taking plant size into account as a covariate, to avoid misleading interpretations about the biodiversity–resistance relationship.
The relationship between biodiversity and ecosystem functioning (BEF) remains a key issue in ecology (Cardinale et al. 2011). Productivity and stability have been shown to increase with plant species richness in terrestrial ecosystems (Tilman, Reich & Knops 2006; Haddad et al. 2011). Most BEF studies to date have focused on grasslands (Cardinale et al. 2011), but evidence in favour of a similar functional relationship in forest ecosystems is accumulating (Vila et al. 2007; Nadrowski, Wirth & Scherer-Lorenzen 2010). However, most of these studies have focused on the producers, with less attention paid to interactions between multiple trophic levels (Scherber et al. 2010). Observations for crop systems have suggested that plant diversity may decrease the rate of herbivory on focal plants through associational resistance to herbivores (Barbosa et al. 2009). Similarly, forest trees have been shown to be less prone to insect damage when growing in mixtures than when in monospecific (pure) forests (Jactel & Brockerhoff 2007).
Two main hypotheses have been proposed to account for associational resistance: the natural enemies and the resource concentration hypotheses (Root 1973). The first hypothesis posits that natural enemies of herbivores (mostly predators and parasitoids) are more abundant (Andow 1991; Bommarco & Banks 2003), more diverse (Castagneyrol & Jactel 2012) and more effective at controlling herbivore populations (Riihimaki et al. 2005; Jactel et al. 2006) in plant communities with a high species richness. According to the resource concentration hypothesis (Root 1973), herbivores are more likely to find their host plant if it is present at high density, because this increases the probability of immigration to and decreases the probability of emigration from monospecific patches (Hambäck, Ågren & Ericson 2000).
There has been considerable debate about the support for the resource concentration hypothesis in the literature (Otway, Hector & Lawton 2005; Heiermann & Schütz 2008; Sholes 2008; Haddad et al. 2009; Björkman et al. 2010; Bañuelos & Kollmann 2011; Plath et al. 2012), and any consideration of this hypothesis should take into account the mechanisms used by herbivores to find their hosts. Host apparency has been defined as the probability of a plant being found by herbivores (Endara & Coley 2011). As most herbivores feed on a limited range of host plants (Barone 1998; Novotny & Basset 2005), greater plant species richness is associated with a higher abundance of nonhost plants for herbivores, and these nonhost plants can decrease host apparency by interfering with the identification of host plants on the basis of visual (Floater & Zalucki 2000; Dulaurent et al. 2012) and chemical (McNair, Gries & Gries 2000; Jactel et al. 2011) cues. Unlike associational resistance, associational susceptibility results in higher levels of herbivory in diverse plant communities than in monocultures (Barbosa et al. 2009). It may result from, for instance, (i) a spillover of herbivores from their preferred to less preferred species, due to resource depletion (White & Whitham 2000) or (ii) from greater fitness in generalist herbivores feeding on a mixture of several host species (Unsicker et al. 2008; Karban et al. 2010). Associational resistance is more likely to occur with specialist herbivores, whereas associational resistance and associational susceptibility have been reported to occur equally frequently for insect generalists, in forests (Jactel & Brockerhoff 2007).
The relative likelihoods of associational resistance and associational susceptibility may depend not only on the breadth of the herbivores' diet, but also on plant species composition. The identity of the plant species assembled in communities determines the ability of herbivores to shift from one host to another. Two host species from the same genus tend to support herbivore communities that are more similar than those feeding on host species from different families (Frenzel & Brandl 2001; Ødegaard, Diserud & Østbye 2005; Novotny et al. 2010). This has important consequences in terms of herbivory, as the loss of fitness experienced by herbivores shifting from one host plant to another increases with the phylogenetic distance between the two host plants (Bertheau et al. 2010). Associational resistance is, therefore, more likely to occur in communities of plants with high phylogenetic diversity (Jactel & Brockerhoff 2007; Pearse & Hipp 2009). Phylogenetic diversity within communities recently emerged as a key component of biodiversity and a better predictor of the BEF relationship than species richness (Cadotte, Cardinale & Oakley 2008; Cadotte et al. 2010). Phylogenetic diversity reflects the entire evolutionary history of species and may therefore be considered as a proxy for functional diversity, which quantifies the similarity between communities on the basis of the functional traits of species (Cadotte, Carscadden & Mirotchnick 2011). Recent studies have suggested that associational susceptibility is more likely to occur within plant communities consisting of assemblages of more closely phylogenetically related species (Jactel & Brockerhoff 2007; Pearse & Hipp 2009; Ness, Rollinson & Whitney 2011; Yguel et al. 2011). However, despite the covariance of functional and phylogenetic diversities, these two factors may have different effects on ecosystem functions, such as productivity (Flynn et al. 2011) and herbivory (Pearse & Hipp 2009).
Until recently, most studies on the diversity–resistance relationship in forests were based on observational data (but see Vehviläinen et al. 2006; Vehviläinejn & Koricheva 2006). However, reports of diversity gradients in forest ecosystems have often failed to take into account confounding factors, such as environmental (e.g. climate) or management (Nadrowski, Wirth & Scherer-Lorenzen 2010) influences. It is therefore increasingly acknowledged that experiments controlling for both species diversity and identity may help to unravel the mechanisms underlying the diversity–resistance relationship. However, even in well-designed experiments, it is impossible to avoid all the confounding factors, defined by Huston (1997) as ‘hidden treatments in ecological experiments’. For example, increases in plant species richness are accompanied by a greater dilution of individual species in substitutive designs so that disentangling the effects of plant diversity per se vs. plant density or apparency remains difficult.
We carried out a large-scale field experiment (12 ha) to test the following hypotheses: (i) tree species diversity favours associational resistance (AR) to forest insects, (ii) the magnitude of AR increases with the functional or phylogenetic diversity of trees, (iii) the effects of tree species richness and composition are stronger for specialist insect herbivores than for generalists. In line with the approaches classically used in biodiversity experiments, we first tested these hypotheses at the plot scale, focusing on pedunculate oaks. Increasing tree diversity increased the probability of association with species growing faster than oaks (i.e. taller trees), due to species-specific differences in growth rate, and this may constitute a hidden treatment. In a second step, we therefore calculated an index of tree apparency, based on the height of individual focal oak trees and their nearest neighbours, and used a statistical modelling approach to test a fourth hypothesis that the effect of tree diversity on insect herbivory is mediated principally by tree apparency.
Materials and methods
This study was carried out 40 km south of Bordeaux (44°440 N, 00°460 W) on the Observatoire Régional de la Phénologie (ORPHEE) experiment, which belongs to the worldwide Tree Diversity Network (TreeDivNet1). The experimental plantation was established in 2008 on a clear cut of former maritime pine stands. Stumps were removed and the site, on a sandy podzol, was ploughed and fertilized with phosphorus and potassium before planting. In total, 25 600 trees of five native species (European birch, Betula pendula; pedunculate oak, Quercus robur; Pyrenean oak, Quercus pyrenaica, holm oak, Quercus ilex; and maritime pine, Pinus pinaster) were planted within a 12 ha area (Fig. 1).
Eight blocks were established, with 32 plots in every block, corresponding to the 31 possible combinations of one to five species, with an additional replicate of the combination of five species. Each plot contained 10 rows of 10 trees planted 2 m apart, resulting in 100 trees per plot, with a plot area of 400 m². Tree species mixtures were established according to a substitutive design, keeping tree density equal across plots. Within plots, individual trees from different species were planted in a regular alternate pattern, such that a tree from a given species had at least one neighbour from each of the other species within a 2 m radius (Fig. 1). The understorey vegetation was mowed once per year. Plots were separated by a distance of three metres and were randomly distributed within blocks. Blocks covered an area of 100 × 175 m, and the entire experimental site was fenced to prevent grazing by mammalian herbivores.
Assessment of herbivory on pedunculate oak
We focused on the 17 plots containing Q. robur (i.e. one plot of each of the 16 possible combinations of one to five species plus an additional five-species mixture plot). Five pedunculate oak saplings were chosen at random from among the 36 innermost trees of the plot. The outermost two rows of each plot were excluded, to avoid edge effects. Herbivory was assessed once per year, in July, by the visual inspection of 20 leaves per tree. Two branches at the top and two branches at the bottom of each sapling were selected at random. At each height, five leaves were sampled at the tip of one branch, and five were sampled from the base of the other branch, to ensure that both young and old leaves were assessed. In total, 425 saplings (5 trees × 17 plots × 5 blocks) were surveyed in 2010 and 2011 (hereafter referred to as focal trees). With the exception of 20 saplings (of 425) that died between 2010 and 2011, herbivory was assessed on the same individuals each year. In 2011, we randomly chose new saplings in the plots where some died to keep the sampling effort constant (i.e. five trees per plot).
Damage by insect herbivores was assigned to three different trophic guilds: leaf chewers (mostly adult Curculionidae or Chrysomelidae, Lepidoptera larvae and grasshoppers), skeletonizers (adult grasshoppers and Tenthredinoidea larvae) and leaf miners (mostly Microlepidoptera). We observed almost no damage due to leaf rollers and gall makers. Damage by skeletonizers and chewers was pooled into a single category (‘leaf chewers’) for analysis because skeletonizing damage was rare and because the same insect species can behave as skeletonizers in early instars and as chewers in later instars (personal observations). We estimated the percentage leaf area removed (LAR) by chewing herbivores for each leaf, using seven classes (0%, 1–5%, 6–15%, 16–25%, 26–50%, 51–75% and >75%) and then averaged the value per sapling, using the median of each class. For the sake of consistency, leaf damage was assessed by the same person in 2010 and 2011 (BC). Previous studies in the studied area (Giffard et al. 2012) showed that most of leaf chewer species are generalist herbivores known to feed on several hosts within different genera or families, including Quercus spp., Betula spp. and even Pinus spp. for some herbivore species (see Table S1 for a list of the most common oak defoliators in the study area).
Leaf miners were identified to genus level on the basis of mine location and shape. The most abundant genera were Phyllonorycter (found on 42.5% of infested saplings), Ectoedemia (42.2%), Stigmella (37.4%), Profenusa (18.3%) and Tisheria (13%). Within these genera, European species were considered oligophagous, because they are able to develop only on a narrow range of Fagacae species. Some can also feed on Castanea sativa, but this tree species was not present in the study area so that leaf miners can be regarded as specialist herbivores. Leaf miner damage was frequent, but affected only a small proportion of the leaf area. We therefore used the total density of leaf mines per focal tree (number of mines per sampled leaves summed at the tree level) to quantify the damage caused by these herbivores.
Quantification of tree species diversity
Effects of tree diversity at the plot scale
Our main objective was to investigate the effects of tree diversity on insect herbivory on pedunculate oak saplings. The experiment was designed for explicit tests of the effect of tree species richness and species assemblages on herbivory. We used eight variables to quantify tree diversity: (i) a categorical variable with each species composition as a category, (ii) the number of tree species planted (species richness), (iii) the proportion of Q. robur, (iv) the proportion of deciduous oak species (Q. robur + Q. pyrenaica), (v) the proportion of all oak species (Q. robur + Q. pyrenaica + Q. ilex), (vi) the proportion of pines, (vii) phylogenetic diversity (PD) and (viii) functional diversity (FD). With the exception of ‘composition’, all variables were treated as continuous variables (Table 1). The proportions of Quercus robur, Q. robur + Q. pyrenaica and oak species were converted into dilution (1 – proportion), in line with our objective of testing the effect of increasing diversity, which results in a greater dilution of host trees. We used variables 3–5 to test the resource concentration hypothesis. The proportion of pine (Variable 6) was used to assess the possible effect of nonhost trees on both leaf chewers and leaf miners.
Table 1. List and type of variables describing tree diversity and tree apparency introduced separately into univariate mixed models
Range, type and number of values
AIC for leaf miners
AIC for leaf chewers
The corresponding Akaike's information criterion (AIC) values are reported for both leaf chewers and leaf miners. AIC values in bold typeface correspond to the best models (with the lowest AIC).
(S) Tree species richness; (Composition) Species composition; (DilutionQr) Dilution of Q. robur; (DilutionQrQp) Dilution of Q. robur + Q. pyrenaica; (DilutionOak) Dilution of Quercus sp.; (ConcentrationPine) Proportion of pines; (PD) Phylogenetic diversity; (FDis) Functional dispersion; (HF) Focal tree height (cm); (ΔHd) Tree apparency.
1–5 (5 different values)
(16 different values)
0.2–1 (5 different values)
0.25–1 (6 different values)
0.33–1 (6 different values)
0–0.5 (5 different values)
0–4.39 (16 different values)
0–2.38 (13 different values)
1–4 (4 different values)
(20 different values)
0–1 (5 different values)
0–1 (7 different values)
0.25–1 (6 different values)
0–0.5 (4 different values)
0–4.39 (20 different values)
0–2.44 (15 different values)
10–181 (400 continuous values)
−107–49 (400 continuous values)
Phylogenetic diversity was used as a measure of both tree species diversity (richness and relative abundance) and evolutionary similarity between associated tree species. Recent studies focusing on the ecological consequences of phylogenetic diversity have highlighted the need to account for the relative abundance of species within communities when calculating phylogenetic diversity (Allen, Kon & Bar-Yam 2009; Cadotte et al. 2010; Ness, Rollinson & Whitney 2011). Abundance-weighted phylogenetic diversity (PD) was therefore calculated with the Phylogenetic Entropy index developed by Allen, Kon & Bar-Yam (2009). This index is derived from Shannon's diversity index and accounts for the relative abundance of species within plots and among local neighbours:
where T is a rooted phylogenetic tree for the community, lb is the length of branch b, and pb is the proportion of individuals in the community ‘descending’ from the branch b (i.e. of all species attached to branch b as ‘leaves’, see Fig. S1).
As the common ancestor of pines and oaks existed long before the common ancestor of oaks, the use of the divergence time from the common ancestor of two species as a measure of branch length would have resulted in PD values very similar to the percentage of pine trees in the mixtures (i.e. outlier effect of pines). We therefore defined branch length on the basis of taxonomic distances (Fig.S1), as defined by Poulin & Mouillot (2003). With this method, PD accounts for the branching pattern of the tree (Faith 1992), with higher values obtained for plots containing pines.
Herbivory patterns may also be accounted for by plant functional diversity. We used the FDis metrics developed by Laliberté & Legendre (2010), corresponding to the mean distance, in multidimensional trait space, of individual species to the centroid of all species. We chose to use this index because it is not affected by species richness and can be calculated for two species mixtures. FDis ranges between zero (all species have the same traits) and one (maximal trait dissimilarity between species). FDis was calculated with the dbFD function in the FD package in R, using traits thought to have a possible effect on the choice and performance of insect herbivores (Agrawal & Fishbein 2006; Jactel et al. 2012): specific leaf area (SLA), leaf area, leaf life span, leaf N content, leaf P content, maximum tree height and resistance to drought (Table S2). We used mean trait values averaged by species from the LEDA database (Kleyer et al. 2008) and those provided by the European BACCARA2 and REINFFORCE3 projects.
Effects of tree diversity at the neighbourhood scale
As we described herbivory at the tree level and measured individual tree height, we were also able to define tree diversity around each individual sampled oak sapling. Hereafter, we use the term ‘neighbourhood’ to refer to the eight closest neighbours of focal trees, located within a 2.83 m radius (2√2). As a result of the regular planting design within plots, tree species compositions were almost identical at the plot and neighbourhood scales, with the exception of the five-species mixtures and some of the four-species mixtures, in which some focal trees had no conspecific neighbour (Fig. 1).
The manipulation of tree diversity had consequences for the vertical structure of plots. Pines and birches grow faster than oak trees. Increases in tree species diversity thus resulted in a greater diversity of tree heights within plots and a more complex vertical structure of plots. We made use of this variability to assess the effect of tree apparency on herbivory. Tree apparency was treated as a proxy for the detectability of the tree to foraging insect herbivores. We used the total height of focal trees (HF) at the end of the growing season as an initial surrogate for tree apparency. However, even tall trees may be poorly apparent if they are surrounded by taller trees, as was the case when oaks were surrounded by birches or pines. We therefore also used the total height of the eight closest neighbours (HNi) to calculate another index of tree apparency (ΔHd) accounting for the mean difference in tree heights between focal and neighbouring trees, weighted by the distance (d) between the focal tree (F) and each of the neighbouring trees (Ni):
Negative values of ΔHd indicated that the focal tree was, on average, smaller (and therefore less apparent) than its neighbours.
We preferred mixed models over anova for assessments of the effects of tree diversity and tree apparency on insect herbivory, because they allow the inclusion of multiple nested random-effect terms (Zuur et al. 2009) to account for spatial (e.g. 17 nonindependent plots within blocks) and temporal autocorrelations (two repeated measures on the same individual trees, in 2010 and 2011).
At the plot scale, response variables [% leaf area removed (LAR) and density of leaf mines] were averaged across the five sampled oak saplings per plot and per year. At the neighbourhood scale, response and explanatory variables were estimated for each focal sapling and individual oaks were therefore used as replicates. Consequently, we declared different random-effect structures for the analyses performed at different scales. At the plot scale, we declared blocks and plots as nested random effects (the effect of year was included within the random effect of plot). At the neighbourhood scale, the declared random effects were Block/Plot/Year (i.e. year nested within plot, nested within block) to account for temporal and spatial autocorrelation between measurements on saplings within the same block, plot and year.
Before analysis, continuous explanatory variables were centred (by subtracting the sample mean from all observations) and reduced (by dividing the centred variables by their sample standard deviation), to obtain model coefficients that were comparable within and between models (Schielzeth 2010). The centring of variables also ensures that the main effects are biologically interpretable, even when involved in interactions (Schielzeth 2010).
The variables describing tree diversity at the plot and neighbourhood scales were not independent, because of the regularity of the design of the ORPHEE experiment. However, the introduction of collinear explanatory variables into multivariate models may lead to inaccurate model parameterization and the exclusion of relevant predictors during model selection (Graham 2003). We addressed the problem of collinearity between variables, by constructing separate univariate models for each explanatory variable for leaf chewers and leaf miners and then calculating Akaike's information criterion (AIC) for each model. This approach to model selection makes it possible to compare models describing the same output variable, but with different explanatory variables (Burnham & Anderson 2002). At the plot scale, we retained the variable that gave the lowest AIC (Table 1) for the modelling of the response of leaf chewers and leaf miners (Burnham & Anderson 2002). At the neighbourhood scale, we also generated separate univariate models for leaf miners and leaf chewers, for variables describing tree diversity and tree apparency. We then used the AIC to identify the variable best accounting for herbivory. We could not compare AIC between models at the plot and neighbourhood scales because these models had different random-effect structures.
The effect of tree apparency may depend on tree diversity. The height of focal trees and tree diversity components were not correlated by construction (see Appendix S1). When focal tree height was identified as the component of tree apparency best describing herbivory, it could be directly tested in interaction with the best explanatory variable for tree diversity identified at the plot scale. Unlike HF, ΔHd was not independent of tree diversity and may be considered as a ‘hidden treatment’ (Fig. 2 and Table S3). Birches and most of the pines were more than twice taller than pedunculate oaks (see Table S4) and the probability of the inclusion of birches and/or pines in the community increased with tree species richness. Tree apparency emerged as the best predictor of leaf miners abundance (see 'Results'). To test a possible interaction between tree apparency and tree diversity effects, we regressed the best tree diversity explanatory variable (i.e. giving the lowest AIC in univariate model) against tree apparency and used the residual as single independent variable (Graham 2003). This method removed the statistical collinearity between covariables and allowed disentangling unique and shared effects of the two explanatory variables on leaf miners abundance (Graham 2003).
Leaf damage by leaf chewers was analysed with the lme procedure (Pinheiro et al. 2012) in R. A logit transformation was applied to proportion data (% LAR), to satisfy the assumptions of statistical tests (Warton & Hui 2010). The total density of leaf miners was analysed with the lmer function in the lme4 R package (Bates, Maechler & Bolker 2011), specifying a Poisson error for count data.
Herbivory response to tree diversity at the plot scale
Herbivory by leaf chewers was not significantly affected by any variable of tree diversity (i.e. tree species assemblages, species richness, dilution, PD and FDis) at the plot scale (Table 1, Fig. 3a). Even the dilution of oak species, which was identified as the best explanatory variable (Table 1), had no significant effect (F1,79 = 1.42, P =0.238, Fig. 3b).
By contrast, the abundance of leaf miners decreased significantly with increasing tree diversity at the plot scale (Fig. 3c), for all variables other than species composition (z =0.39, P =0.699). The dilution of deciduous oak species (i.e. Quercus robur + Q. pyrenaica) was the best explanatory variable (z =−3.34, P <0.001, Table 1 and Fig. 3d). Thus, the mean abundance of leaf miners per plot decreased significantly with tree diversity, due to the greater dilution of host trees in more diverse plots.
Herbivory response to tree diversity, height and apparency at the neighbourhood scale
The insect herbivory response to tree diversity variables at the neighbourhood scale was similar to that observed at the plot scale. For leaf chewers, none of the eight diversity variables had a significant effect (Fig. 4a). By contrast, the abundance of leaf miners decreased significantly with increasing tree diversity, for all diversity variables other than tree species composition (Fig. 4c).
For leaf chewers, both tree height (F1,676 = 41.91, P <0.001) and tree apparency (i.e. ΔHd, F1,676 = 8.12, P =0.004) had a significant positive effect on herbivory, with taller and more apparent saplings being more prone to damage than less apparent ones (Fig. 4b). Variable selection on the basis of AIC identified the height of focal tree as the best explanatory variable for herbivory at the local scale (Table 1). We introduced each of the best variables selected on the basis of AIC comparisons (i.e. dilution of oak species at the plot scale and individual sapling height) into a single model to test the interaction between diversity and apparency effects. There was no significant interaction between tree height and oak dilution at the plot scale (F1,675 < 0.01, P =0.990). Thus, taller saplings consistently experienced more chewing damage than smaller saplings, regardless of the diversity of the surrounding trees (Fig. 4b).
Model comparisons based on AIC identified ΔHd as the variable best accounting for the abundance of leaf miners at the neighbourhood scale (Table 1). Leaf miner infestation increased significantly with increasing oak sapling apparency (z =6.31, P <0.001 Fig. 4d). This suggests that the abundance of leaf miners was not related to tree diversity per se, but to the local structural heterogeneity generated by tree diversity.
Tree apparency (ΔHd) and host (Q. robur + Q. pyrenaica) dilution were highly correlated (Pearson's r = 0.48, P <0.001, Fig. 2). We therefore regressed the dilution of deciduous oak species at the plot scale against ΔHd and used the residuals to test the interaction effect between the two explanatory variables (Graham 2003). There was a significant interaction between tree apparency and host dilution at the plot scale (z = −2.36, P = 0.018). However, when the correlation between the two explanatory variables was accounted for, the ‘pure’ effect of host concentration was no longer significant (z = −4.43, P =0.669). The negative coefficient for the interaction term (−0.11 ± 0.05) indicates that the magnitude of the effect of tree apparency on the abundance of leaf miners was stronger in monocultures and decreased as host dilution increased.
We show here that tree diversity triggers different responses from chewing and mining insect herbivores. Increasing tree diversity was associated with a significant increase in associational resistance to leaf miners, but had no effect on resistance to leaf chewers. However, the key result of this study was the finding that the observed effect of tree diversity on leaf miners could mostly be explained by the differences in tree height resulting from the manipulation of tree diversity. We also found that chewing herbivores caused more damage on taller saplings. For both chewing and mining insects, host plant size therefore emerged as a key driver of herbivory.
Greater apparency results in higher levels of herbivory
Plant apparency is broadly defined as the likelihood of a plant being found by herbivores (Feeny 1970; Endara & Coley 2011). Apparency can be broken down into two components: plant life span (as long-living species, trees are more likely to be found by herbivores than annual plants) and plant accessibility. In this study, we focused on accessibility and quantified tree apparency as the mean difference between the height of an individual tree and that of its closest neighbours. We calculated this difference because even tall trees may be poorly apparent if surrounded by taller neighbours.
The abundance of leaf miners increased significantly with the size of oak saplings relative to their neighbours (i.e. host apparency), consistent with the findings of previous studies reporting a lower infestation of host trees concealed by nonhost plants (Floater & Zalucki 2000; Hughes 2012), probably because nonhost plants can disrupt visual (Dulaurent et al. 2012) or olfactory (Jactel et al. 2011) cues for host localization. Leaf miners are generally distributed in a nonrandom manner between and within trees, suggesting an active selection of oviposition sites by females (Cornelissen & Stiling 2006). Tree diversity may therefore have resulted in associational resistance to these specialist herbivores by limiting their capacity to find hosts. In contrast, damage due to chewing herbivores was positively correlated with, and best explained by, the absolute height of individual host trees, rather than by the difference in size between focal trees and their neighbours. According to the ‘appropriate landing hypothesis’ (Finch & Collier 2000), host selection by insects is a sequential process involving an initial attraction at distance based on chemical and visual cues and then a ‘should I stay or should I go?’ decision based on cues providing information about food quality (Finch & Collier 2000). The differences between the effects of tree apparency and tree height on damage by chewing herbivores may reflect the relative importance of host finding (tree apparency) and host acceptance (tree height) mechanisms for these herbivores: the taller saplings may have been easier to reach (Floater & Zalucki 2000), and they may have provided herbivores with more suitable feeding resources (Lawton 1983; Herms & Mattson 1992). Indeed, according to the growth-differentiation balance hypothesis, plants cannot simultaneously allocate resources to growth and defence (Herms & Mattson 1992). Thus, faster growing individuals may have weaker defences, with the production of fewer secondary metabolites, for example, resulting in higher levels of herbivory. Consistent with this, herbivore density on plants has been shown to be higher on faster growing plants, in accordance with the so-called ‘plant vigour hypothesis’ (Cornelissen, Fernandes & Vasconcellos-Neto 2008). In addition, we observed in our experimental site that oak saplings produced up to four successive generations of leaves during the growing season and that taller saplings displayed more regrowth than shorter saplings. One possible explanation for the positive correlation between sapling height and herbivory may therefore be the provision of larger numbers of young leaves by taller saplings (Lawton 1983), younger leaves being more palatable because of their high water and nitrogen contents and low tannin concentrations (Murakami & Wada 1997; Tikkanen et al. 2003; Murakami et al. 2005).
We cannot exclude that changes in herbivory with oak apparency may also have been driven by changes in plant quality. Taller oaks may have received more light than oak saplings surrounded by taller neighbours. Leaves exposed to direct light generally have higher concentrations of defence compounds and lower nitrogen content (Dudt & Shure 1994), which make them less palatable for insect herbivores. By contrast, Barber & Marquis (2011) reported higher levels of insect herbivory on saplings previously exposed to high-light intensity and argued that the search for oviposition sites may have led females to choose saplings with more foliage. Neighbouring plants may have also changed microclimate conditions that could have adversely affected insect herbivores (Barbosa et al. 2009).
Tree diversity and resistance to herbivores: beyond the paradigm
We found no significant effect of tree diversity per se on damage due to chewing herbivores, whereas tree diversity triggered associational resistance to leaf miners. Yet, in our study site, most of leaf chewers were generalist herbivores, while leaf miners found on oak are specialists on genus Quercus (Giffard et al. 2012). Our findings are therefore consistent with the results of the meta-analysis by Jactel & Brockerhoff (2007), which showed consistently lower levels of damage due to oligophagous insects in more diverse forests and a more variable effect of tree diversity on polyphagous herbivores. Other studies have reported oak trees to be more prone to damage in mixed than in pure stands (Vehviläinen, Koricheva & Ruohomäki 2007). There are two main reasons for these discrepancies between our results and those of previous studies: (i) we considered the damage caused by the whole community of free-feeding herbivores rather than by a single herbivore species and (ii) we focused on saplings rather than mature trees.
Most previous studies focused on the damage caused by one or few particular herbivore species (Jactel et al. 2006; Heiermann & Schütz 2008). This approach can be used to identify the mechanisms involved in the diversity–resistance relationship (Jactel & Brockerhoff 2007; Dulaurent et al. 2012), but is not the most relevant for estimating the effects of tree diversity on resistance to herbivores from the plant perspective, because several different herbivore species may feed on a single host tree (Southwood et al. 2004; Wielgoss et al. 2012). Herbivores feeding on the same plant may or may not interact, and direct and indirect interactions may range from competition via the removal of biomass or defence induction to facilitation through physiological weakening or the induction of new flushes of growth (Kaplan & Denno 2007). Consequently, plant diversity may have opposite effects on different herbivore species (Plath et al. 2012), resulting in an unpredictable overall effect of plant diversity on total herbivory. Indeed, if the entire community of herbivorous insects is considered, tree diversity may have no effect on herbivory damage (Schuldt et al. 2010; Plath et al. 2011) or effects ranging from associational susceptibility(Schuldt et al. 2010) to associational resistance (Plath et al. 2011), depending on tree species considered. The diversity–resistance hypothesis has received little unequivocal support to date from experiments comparing primary productivity in the presence or absence of herbivores, along a gradient of plant species richness (Cardinale et al. 2011).
Because of the size of the plots (20 × 20 m²), the ORPHEE experiment may be currently more suitable to address interactions between tree diversity and herbivory at a local (i.e. neighbourhood) scale, and caution is needed in extrapolating the outcomes of our study to the forest stand scale. Furthermore, trees are not herbs, and forests are long-living ecosystems in which herbivore communities associated with particular host trees change with ecological succession (Jeffries, Marquis & Forkner 2006) and host ontogeny (Lawton 1983; Campos et al. 2006; Thomas, Sztaba & Smith 2011). As plant resistance to herbivores also changes with plant ontogeny (Boege & Marquis 2005; Barton & Koricheva 2010; Boege, Barton & Dirzo 2011), different mechanisms may also be involved in the diversity–resistance relationship in seedlings and in adult trees. The absence of an effect of tree diversity on herbivory by leaf chewers in our study does not call into question the validity of the diversity–resistance relationship for these insects. Instead, it highlights the need for long-term experiments, to improve our understanding of the complex interactions between tree diversity and resistance to herbivores.
Which component of tree diversity is relevant to associational resistance?
Seminal studies focusing on the biodiversity–ecosystem functioning (BEF) relationship used two metrics to describe plant diversity: species richness and the number of functional groups (Siemann et al. 1998; Hector et al. 1999; Tilman, Reich & Knops 2006). Recent developments in this field of research have highlighted the need to account for the identity of associated species. This can be achieved with the functional diversity (FD) and phylogenetic diversity (PD) indices (Cadotte et al. 2009; Cadotte, Carscadden & Mirotchnick 2011). For example, insect herbivory on oaks has been shown to decrease with increasing phylogenetic distance between oaks and their neighbours (Pearse & Hipp 2009; Yguel et al. 2011). This result is consistent with the greater resistance of mixed forest stands associating coniferous and broadleaved trees than of stands of purely coniferous or broadleaved species reported by Jactel & Brockerhoff (2007). However, we were unable to confirm the key role of phylogenetic diversity in associational resistance in our experiment, as this factor was found to be a slightly less relevant predictor of the abundance of leaf miners than host dilution.
The significant effect of host dilution on the abundance of leaf miners is consistent with the ‘resource concentration hypothesis’, according to which, specialist herbivores concentrate in patches in which their resource is abundant, because they have a higher probability of immigrating into and a lower probability of emigrating from such patches (Root 1973; Hambäck, Ågren & Ericson 2000). However, only limited support has been obtained for this prediction (Heiermann & Schütz 2008; Björkman et al. 2010; Giffard et al. 2012). For example, Bañuelos & Kollmann (2011) found that the abundance of Phytomyza ilicis, a leaf miner feeding on holly trees (Ilex aquifolium), decreased with increasing host density.
We also detected a ‘hidden treatment’ that could modify interpretation of the diversity effect. In our experiment, tree apparency covaried with the diversity of the tree species assemblage, due to differences in growth rate between species. Birches and pines grew more rapidly than oaks, so the apparency of oak saplings may have depended on the presence or absence of birches and/or pines in the neighbourhood. The probability of faster growing species being neighbours of pedunculate oaks increased with tree diversity, resulting in a possible hidden treatment underlying the effect of tree diversity. This finding is analogous to the demonstration of a ‘dominance’ or ‘sampling’ effect in BEF studies, in which increasing plant diversity resulted in a higher probability of incorporating plant with a dominant effect on ecosystem processes (Huston 1997; Loreau & Hector 2001). In our case, the sampling effect relates to the incorporation of rapidly growing species in tree assemblages.
Our experimental study provides new evidence for the existence of a relationship between diversity and resistance, with lower levels of damage by leaf miners in more diverse tree species assemblages. However, the main contribution of this work was the breaking down of the diversity effect mainly into two interacting mechanisms acting at two different spatial scales. The first of these mechanisms is based on host dilution among nonhost plants, which increases with increasing tree species richness. The second is based on tree apparency being lower due to the presence of taller trees, when the nonhost trees grow more rapidly than the host trees. Both mechanisms may have resulted in the disruption of host location and colonization, and both are dependent on the presence of particular nonhost species (sampling effect). In addition, the above two mechanisms (host dilution and tree apparency) did interact. Leaf miners responded more strongly to tree apparency when their host species were concentrated but became less selective about host tree size when their feeding resource was diluted and more difficult to find, suggesting changes in host searching patterns depending on local resource availability (Wang et al. 2010).
These findings highlight the fact that insufficient consideration of plant size as a covariate may lead to misleading interpretations about the existence of such effects of biodiversity, particularly during forest regeneration. They also have implications for the design of new planted forests, which are mostly managed as monocultures. The complementation between tree species grown for wood production and more rapidly growing pioneer species may provide effective protection against pest insects through visual or olfactory disruption of host finding while producing additional biomass. However, long-term studies are required to determine the consequences of decreasing herbivory for forest productivity throughout the entire forest rotation.
We would like to thank the experimental unit at INRA Pierroton for establishing and maintaining the ORPHEE experiment. We thank François Bizet and Guillaume Cacadavid for his valuable help during field work. Our thanks also go to Pr. Julia Koricheva and two anonymous reviewers for their comments that contributed to improve the quality of the manuscript. The research reported here was conducted as part of the European BACCARA and FunDiv projects, which received funding from the European Commission's Seventh Framework Programme (FP7/2007-2013), under grant agreement nos. 226299 and 265171 respectively.