Design of species assemblages with contrasting species and functional diversities
We designed eight assemblages of native and perennial plants differing in terms of species richness (three levels), functional diversity of the traits involved in plant–arthropod interactions (two levels) and species identity (two sets of species). We combined these first two factors to define four categories of plant assemblages for further study:
Low functional diversity and medium species richness (14 species), LFMS;
High functional diversity and low species richness (9 species), HFLS;
High functional diversity and medium species richness (14 species), HFMS;
High functional diversity and high species richness (29 species), HFHS.
For each of these four categories, we designed two assemblages with different species identities, as described in the Supplementary information, resulting in eight plant assemblages in total. Functional characterization was based on a rough classification of plant species into functional groups (Supplementary Table S1), according to the mains traits involved in plant–species interactions easily accessible from databases: (1) flower resources, i.e. floral and extrafloral nectar or pollen, (2) accessibility of the resource, depending on flower shape, (3) availability of the resource, i.e. the flowering period and (4) flowering height.
We generated the seed mixtures from commercial seeds, using ecotypes of local origin wherever possible (northern part of the Parisian basin, France). All applicable international, national, and institutional guidelines relevant for the use of plants were followed.
Experimental design
The experiment was conducted between 2013 and 2017 in a 6.5-ha field at Grignon, France (N 48.837, E 1.956), on a deep loamy clay soil, in which soil depth decreased along a gradient from north to south. The field was divided in three blocks running from north to south to take this soil heterogeneity into account.
Each assemblage was sown on a 6 × 44 m2 strip, with three replicates (Supplementary Fig. S2), with each assemblage represented once per block. A control treatment, sown with the same crop species as the rest of the field, was also included in the experimental design, resulting in nine experimental treatments in total. From the autumn of 2013 to the 2017 harvest, a winter barley–maize–faba bean–oilseed rape rotation was grown in the field. Crops were managed without insecticide treatment, but with a mean of 0.75 fungicide and 1.25 herbicide treatments per year. The observations were made in faba bean in 2016 and in oilseed rape in 2017.
Botanical assessments and functional characterization of the plant communities
Botanical assessments were conducted in April and June, in 2016 and 2017. In each treatment, the vegetation was assessed in 3 × 15 m2 plots at a position representative of the whole strip, generally in the center of the strip, to prevent edge effects. The percentage of the ground covered by each sown or spontaneously growing plant species was estimated by eye, by the same observer in each case. We noted the phenological development stage of each species in each treatment on an 11-point scale, to ensure an accurate assessment of flowering phenology. In the control plots (sown with the crop species only), we took into account the resources provided by weed species.
The functional characterization of plant communities was based on the plant traits assumed to be involved in plant–parasitoid interactions6 (Supplementary Table S3). These traits were related to (1) the provision of trophic resources (presence of floral and extrafloral nectar, qualitative estimation of floral nectar), (2) the temporal availability of the resource (date of flowering onset and duration of flowering), (3) flower attractiveness (flower or inflorescence diameter, color, UV reflectance pattern), (4) nectar accessibility (flower opening diameter, corolla height, nectar depth and nectar tube diameter) and (5) the provision of physical habitats (leaf distribution, vegetative and flowering height). We measured most of these traits, particularly all those relating to flower morphology, phenology and nectar provision (see more detailed methods in the Supplementary information). Only a few were retrieved from previous publications and online databases: flower color and UV reflectance pattern, leaf distribution, vegetative and flower height.
These traits were used (1) to determine the accessibility of nectar to each parasitoid (see below) and (2) to calculate the functional diversity of the plant assemblages. We calculated functional dispersion as the abundance-weighted mean distance of individual species from the centroid of all species in the trait space50 and Rao quadratic entropy51. Since these two parameters were highly correlated (Supplementary information), we considered only functional dispersion a measurement of functional diversity. The traits associated with the provision, availability and accessibility of nectar resources were measured for all the dicotyledonous species sown and for all spontaneous species occurring in the plant communities and flowering during parasitoid activity. Overall, considering the traits we measured and those retrieved from databases, the trait matrix was complete for more than 95% of the species, accounting for 99.6% of total plant cover.
Assessment of the levels of parasitism on five herbivorous pests of faba bean and oilseed rape
In the adjacent crop, 5 and 20 m from the wildflower strip, we measured the level of parasitism in one herbivorous pest of faba bean (2016) and four herbivorous pests of oilseed rape (2017). We chose a distance close to the strip (5 m) to prevent confounding effects with the other adjacent strips, knowing that their effect is the strongest in the first few meters from the strip52. A further distance was also chosen (20 m) to determine whether the strips promoted biological control at field level, while taking into account the spatial constraint of the distance between strips (50 m between opposing strips).
All the protocols are detailed in the Supplementary information. Parasitism was assessed in Bruchus rufimanus larvae after the visual examination of faba bean seeds after harvest. For oilseed rape, we collected and reared Ceutorhynchus pallidactylus and Psylliodes chrysocephala larvae until the adult stage or parasitoid emergence. In Brassicogethes aeneus larvae, parasitism was assessed by observing the eggs of Tersilochus heterocerus in the host larvae in oilseed rape flowers. Finally, after oilseed rape harvest, we retrieved cocoons of Dasineura brassicae from the soil, which we dissected, recording the number of cocoons occupied by parasitoids.
Measurement of parasitoid traits
We carried out morphological measurements on parasitoids (Supplementary Table S4), to determine their degree of access to the nectar provided by plants, as a function of the size of their mouthparts and head, which limit corolla penetration, using an approach analogous to that of van Rijn and Wäckers16. Parasitoid individuals, preserved in 70% ethanol, were obtained (1) from our rearing experiments (for Bruchus rufimanus, Psylliodes chrysocephala and Ceutorhynchus pallidactylus), (2) from the dissection of cocoons for Dasineura brassicae or (3) by field sampling in the flower strips with a sweep net in April 2017 to collect Tersilochus heterocerus, parasitoids of Brassicogethes aeneus identified with53. For each parasitoid species or morphospecies, we measured, on at least 10 individuals, proboscis length, proboscis width (at mid-length)54 and the maximum dorsal head width, including the eyes. Observations were carried out under a binocular microscope (Leica M80, 60 ×) linked to a video camera (Moticam 10, Motic), and measurements were made with ImageJ v1.50i digital image analysis software (National Institute of Health, Bethesda, http://imagej.nih.gov/ij).
Nectar resources for parasitoids
We estimated the amount of nectar provided by the plants by summing, for each flower strip corresponding to a treatment, the percent cover of plants providing available and accessible nectar, as assessed in vegetation surveys. Separate estimates were obtained for each parasitoid species or morphospecies.
Plant species producing floral or extrafloral nectar were first selected on the basis of the observations detailed in the botanical assessment section. Nectar was considered to be available when it was produced during the period of parasitoid activity (Supplementary Table S4), by selecting species at the flowering stage or producing extrafloral nectar based on the phenological observations carried out during the botanical assessments. Nectar accessibility depended on morphological matching between plants and insects. Extrafloral nectar, which is not enclosed in a perianth, but produced on bracts or stipules, was considered to be accessible. We determined the accessibility of floral nectar with a mechanistic trait-based approach (Supplementary Information), by adapting the geometric model proposed by van Rijn and Wäckers16. A decision tree was built (Fig. 2) to take into account the three constraints limiting nectar accessibility: (1) ability of the insect to penetrate the flower, which is dependent on head size and flower opening, (2) ability to reach the nectar, which depends on proboscis length, nectar depth and corolla height, and (3) proboscis width and nectar tube diameter in the presence of nectar.
Statistical analyses
We investigated the effects of the different plant assemblages on the rates of parasitism for the five herbivorous species, at 5 and 20 m from the flower strip, considered separately as individual response variables. We first tested the effect of each assemblage (nine treatments as factors) on parasitism rates. We used generalized linear mixed models in the lme package55, with a binomial error distribution. The models included plot (n = 9 flower strips × 3 replicates = 27), strip (1–3) or block (1–3) as a random effect. All models were run three times with each random effect variable, and the model giving the lowest AIC was retained. Strips consistently yielded the lowest AIC. This factor was therefore introduced as a random effect variable for all statistical analyses. The significance of the fixed effects was evaluated by type II analyses of deviance with Wald chi-squared tests from the Anova function from the car package56. If a significant effect (p value < 0.05) was detected, we performed Tukey-HSD post hoc tests for pairwise multiple comparisons of the estimated marginal means of each treatment with the multcomp package57.
We investigated the effects of the plant assemblages on parasitism rates, by characterizing three explanatory variables: (1) plant species richness, (2) percentage of plant cover providing nectar accessible to parasitoids, estimated from the traits relating to nectar availability and accessibility and (3) functional diversity of all the previously described plant traits involved in plant–insect interactions, including attractiveness. These explanatory variables were standardized to account for the large differences in scale between them. We tested our hypotheses with generalized linear mixed models, assuming that the errors followed a binomial distribution. For each response variable (parasitism on each species at each distance), we tested models including, as fixed effects, these three explanatory variables with their linear and quadratic terms (to test for non-linear patterns), and with two-level interactions between the three linear terms. As described above, the strip was introduced as a random effect variable. Using a multimodel inference procedure58, we tested models including all possible additive combinations of the nine predictors, ranked according to the Akaike information criterion (AIC), fitted by maximum likelihood methods. Models with a ΔAIC < 2 were selected and we present the statistical results for the conditional averaged model. Akaike weights were calculated to assess the relative importance of each predictor. Only predictors with a weight above 0.70 are interpreted.
We tested the validity of the trait-matching approach, by constructing a neutral model in which interactions were randomly selected rather than determined by trait-matching, whilst keeping constant the total number of plant–parasitoid interactions. We tested (1) the morphological trait-match, by randomly selecting the interacting plant species from those that were in flower during the period of adult parasitoid activity and (2) the temporal and morphological trait-match by randomly selecting interacting plant species from all those present in the plant community, whatever their phenological stage. We then summed, for each treatment, the percent cover of these randomly selected plant species. This new explanatory variable was introduced as an alternative to the cover of plants producing accessible nectar in the statistical models investigating the effects of plant assemblages on the parasitism rates for each herbivore. One thousand iterations were performed for each parasitoid. We averaged the AIC of neutral models and compared them with the trait-matching model.
Finally, we assessed the ability of the assemblages to contribute to multi-species parasitism, by using the multiple-threshold approach to determine whether high levels of parasitism could be simultaneously achieved for several different herbivores31. Rather than averaging non-substitutable parasitism rates, the threshold-based approach31 involves the calculation, for each treatment, of the number of herbivores for which the parasitism rate exceeds a given threshold (% of the highest observed parasitism rate for each herbivore). Defining a specific threshold for determining whether a given parasitism rate is sufficiently high to contribute to herbivore parasitism would be arbitrary. We therefore calculated multi-species parasitism rates for a full range of thresholds, from 10 to 90% of the maximum parasitism rate (mean of the three highest recorded values), in 10% steps, and we took this threshold effect into account in the regression analyses. In each experimental treatment and for each threshold, the multi-species parasitism rate was calculated as the number of herbivore parasitism rates locally exceeding the threshold value. Diversity effects might also be expected to be stronger for higher proportions of parasitism rates exceeding the threshold value, to reach multifunctionality, i.e. a larger number of plant species or plant functions would be required to maintain several parasitism rates at high levels31. We therefore included in the regressions an interaction between the threshold variable and the other variables describing plant communities.
We investigated the effect of the plant assemblages on multi-species parasitism, by first testing the effect of each assemblage separately. Visual examination of the results (Fig. 4) showed that multi-species parasitism was dependent on the chosen threshold: when it was very high, only a few assemblages provided a high parasitism rate (e.g. 90% of the maximum observed value) for all herbivores, whereas a threshold of 50% better discriminated between assemblages. We therefore included in the models the threshold variable (continuous) and its interactions with the other fixed-effect variables. We used generalized linear mixed models with a Poisson error distribution and the strip as a random effect. We assessed the significance of effects as previously described for individual parasitism rates. We then tested the relationships between multi-species parasitism and (1) plant species richness (averaged over the two years), (2) percentage of plant cover providing accessible nectar (averaged for all parasitoids) and (3) functional diversity (averaged over the two years). We used the same approach as previously described for individual parasitism rates. The models included an additional effect of threshold values, introduced as a continuous fixed-effect variable, alone and in interaction with all explanatory variables. We prevented over-parameterization, by applying the model selection procedure only to models containing a maximum of five fixed-effect variables.
We estimated the variance explained by the models with the marginal and conditional pseudo-R2 statistic59. Diagnostic residual plots for all models were confirmed with the DHARMa package30. All statistical analyses were performed with R software, version 3.6.360.
Source: Ecology - nature.com
